WO2013070803A1 - Système de stockage de données primaires doté d'une hiérarchisation de données - Google Patents

Système de stockage de données primaires doté d'une hiérarchisation de données Download PDF

Info

Publication number
WO2013070803A1
WO2013070803A1 PCT/US2012/063995 US2012063995W WO2013070803A1 WO 2013070803 A1 WO2013070803 A1 WO 2013070803A1 US 2012063995 W US2012063995 W US 2012063995W WO 2013070803 A1 WO2013070803 A1 WO 2013070803A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
filter
iob
tier
block
Prior art date
Application number
PCT/US2012/063995
Other languages
English (en)
Inventor
Paul A. ASHMORE
Kelly E. LONG
Sebastian Piotr SOBOLEWSKI
Original Assignee
Nexgen Storage, Inc.
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nexgen Storage, Inc. filed Critical Nexgen Storage, Inc.
Publication of WO2013070803A1 publication Critical patent/WO2013070803A1/fr

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F12/00—Accessing, addressing or allocating within memory systems or architectures
    • G06F12/16—Protection against loss of memory contents
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/34—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment
    • G06F11/3409—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment for performance assessment
    • G06F11/3433—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment for performance assessment for load management
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/34—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment
    • G06F11/3466—Performance evaluation by tracing or monitoring
    • G06F11/3485—Performance evaluation by tracing or monitoring for I/O devices
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
    • G06F3/06—Digital input from, or digital output to, record carriers, e.g. RAID, emulated record carriers or networked record carriers
    • G06F3/0601—Interfaces specially adapted for storage systems
    • G06F3/0602—Interfaces specially adapted for storage systems specifically adapted to achieve a particular effect
    • G06F3/0608—Saving storage space on storage systems
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
    • G06F3/06—Digital input from, or digital output to, record carriers, e.g. RAID, emulated record carriers or networked record carriers
    • G06F3/0601—Interfaces specially adapted for storage systems
    • G06F3/0602—Interfaces specially adapted for storage systems specifically adapted to achieve a particular effect
    • G06F3/061—Improving I/O performance
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
    • G06F3/06—Digital input from, or digital output to, record carriers, e.g. RAID, emulated record carriers or networked record carriers
    • G06F3/0601—Interfaces specially adapted for storage systems
    • G06F3/0602—Interfaces specially adapted for storage systems specifically adapted to achieve a particular effect
    • G06F3/0614—Improving the reliability of storage systems
    • G06F3/0619—Improving the reliability of storage systems in relation to data integrity, e.g. data losses, bit errors
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
    • G06F3/06—Digital input from, or digital output to, record carriers, e.g. RAID, emulated record carriers or networked record carriers
    • G06F3/0601—Interfaces specially adapted for storage systems
    • G06F3/0628—Interfaces specially adapted for storage systems making use of a particular technique
    • G06F3/0638—Organizing or formatting or addressing of data
    • G06F3/064—Management of blocks
    • G06F3/0641—De-duplication techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
    • G06F3/06—Digital input from, or digital output to, record carriers, e.g. RAID, emulated record carriers or networked record carriers
    • G06F3/0601—Interfaces specially adapted for storage systems
    • G06F3/0628—Interfaces specially adapted for storage systems making use of a particular technique
    • G06F3/0646—Horizontal data movement in storage systems, i.e. moving data in between storage devices or systems
    • G06F3/0647—Migration mechanisms
    • G06F3/0649—Lifecycle management
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
    • G06F3/06—Digital input from, or digital output to, record carriers, e.g. RAID, emulated record carriers or networked record carriers
    • G06F3/0601—Interfaces specially adapted for storage systems
    • G06F3/0668—Interfaces specially adapted for storage systems adopting a particular infrastructure
    • G06F3/0671—In-line storage system
    • G06F3/0683—Plurality of storage devices
    • G06F3/0685—Hybrid storage combining heterogeneous device types, e.g. hierarchical storage, hybrid arrays
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/07—Responding to the occurrence of a fault, e.g. fault tolerance
    • G06F11/16—Error detection or correction of the data by redundancy in hardware
    • G06F11/20—Error detection or correction of the data by redundancy in hardware using active fault-masking, e.g. by switching out faulty elements or by switching in spare elements
    • G06F11/2002—Error detection or correction of the data by redundancy in hardware using active fault-masking, e.g. by switching out faulty elements or by switching in spare elements where interconnections or communication control functionality are redundant
    • G06F11/2005—Error detection or correction of the data by redundancy in hardware using active fault-masking, e.g. by switching out faulty elements or by switching in spare elements where interconnections or communication control functionality are redundant using redundant communication controllers
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/07—Responding to the occurrence of a fault, e.g. fault tolerance
    • G06F11/16—Error detection or correction of the data by redundancy in hardware
    • G06F11/20—Error detection or correction of the data by redundancy in hardware using active fault-masking, e.g. by switching out faulty elements or by switching in spare elements
    • G06F11/2053—Error detection or correction of the data by redundancy in hardware using active fault-masking, e.g. by switching out faulty elements or by switching in spare elements where persistent mass storage functionality or persistent mass storage control functionality is redundant
    • G06F11/2089—Redundant storage control functionality
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/07—Responding to the occurrence of a fault, e.g. fault tolerance
    • G06F11/16—Error detection or correction of the data by redundancy in hardware
    • G06F11/20—Error detection or correction of the data by redundancy in hardware using active fault-masking, e.g. by switching out faulty elements or by switching in spare elements
    • G06F11/2053—Error detection or correction of the data by redundancy in hardware using active fault-masking, e.g. by switching out faulty elements or by switching in spare elements where persistent mass storage functionality or persistent mass storage control functionality is redundant
    • G06F11/2094—Redundant storage or storage space
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/34—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment
    • G06F11/3409—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment for performance assessment
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2201/00—Indexing scheme relating to error detection, to error correction, and to monitoring
    • G06F2201/88—Monitoring involving counting

Definitions

  • the present invention relates to a primary data storage system suitable for use in a computer network.
  • a computer network is typically comprised of multiple user computers, a primary data storage system that stores data provided by the user computers and provides previously stored data to the user computers, and a network system that facilitates the transfer of data between the user computers and the primary data storage system.
  • the user computers typically have local data storage capacity.
  • the primary data storage system is separate from the user computers with local data storage capacity and provides the ability for the user computers to share data/information with one another.
  • the network system that is between the user computers and the primary data storage system can take a number of forms. For example, there can be a dedicated channel between each of the user computers and the primary data storage system.
  • the network system includes switches (fabric switches) and servers (in certain situations known as initiators) that cooperate to transfer data between the primary data storage system and the user computers.
  • switches fabric switches
  • servers in certain situations known as initiators
  • secondary data storage system provides secondary storage of data, i.e., storage that is not constantly available for use by one or more user computers when the computer network is in a normal/acceptable operating mode.
  • many secondary data storage systems are employed to backup data and to facilitate other maintenance functions.
  • primary data storages are substantially constantly available for use by one or more user computers when the computer network is in a normal/acceptable operating mode that involves substantial interaction with the user computers.
  • the present invention is directed to a primary data storage system comprised of: (a) one or more i/o ports, each i/o port capable of receiving a packet with a block command and providing a packet with a reply, (b) a data store system having at least one data store capable of receiving and storing data in response to a write block command and/or retrieving and providing data in response to a read block command, and (c) a storage processor with a processor and application memory for executing computer code related to the transfer of data between the one or more i/o ports and the at least one data store.
  • the primary data storage system includes a data store system with two or more tiers.
  • Each tiers is comprised of a data store (e.g., a disk drive) or a group of stores that have similar performance characteristics.
  • Each of the tiers has characteristics that distinguish the tier from other tiers. For example, one tier may have low latency (i.e., capable of responding quickly) and low capacity, while another tier may have high latency and high capacity.
  • the storage processor includes a tiering processor that operates to implement tiering functionality in which a block of data or a group of data blocks are stored on whichever tier in the system is the better match to the access pattern of the block or group of blocks, i.e., the pattern in which the block/blocks are being written and/or read.
  • the tiering processor is capable of copying one or more blocks of data from a first tier to a second tier. Because the block or blocks are copied and not moved, there are two copies of the block/blocks, one copy on the first tier and another copy on the second tier. As such, the copying provides data redundancy, i.e., one copy could be corrupted in some fashion and the other copy still used.
  • the tiering processor is capable of removing or overwriting one or more of the blocks on the first tier. In this case, the first tier has a subset of the blocks that are now on the second tier.
  • the pattern of usage of the subset of blocks may dictate that the subset of blocks is a better match to one tier and the pattern of usage of the larger set of blocks other than the subset of blocks is a better match for a different tier.
  • the tiering processor is capable of recognizing that there are two copies of a data block(s) with one copy in one tier and the other copy in another tier and then, if space is needed for other data, overwrite one of the copies.
  • Another embodiment of the primary data storage system employs a tiering processor that is capable of processing a write block command so as to identify one tier of multiple tiers in the system as having the characteristics least likely relative to the other tiers to require, if the data associated with the command were placed on the identified tier, that the data be moved to other tiers. While there is the possibility that the data placed on the identified tier would need to be moved at some point later in time, the tiering processor endeavors to reduce the possibility of such a movement occurring. If all other factors are equal, this leads to the more efficient use of the system by reducing the time spent by the system in moving data from one location to another in the system and increasing the amount of time available to perform other tasks needed by the users of the system.
  • the identification of the tier that is least likely to require movement of the data is based upon attributes of the write data command and related data, characteristics of the tiers, and other statistical information relating to the system.
  • At least two tiers in the system having different atomic blocks sizes is the smallest size of block that the store or stores that comprise the tier process. Typically, this atomic block size is the size of block that the store or stores process most effectively. For example, a particular disk drive may be capable of processing blocks that are 4KB in length but may perform significantly better on blocks that are only 512B in length. In this case, the atomic blocks size is 512B.
  • a block or blocks of interest can be directed to the tier with the atomic block size that most closely matches the size of the block or blocks of interest.
  • the block of interest is stored on the tier with the 4KB atomic block size because accessing that block of interest from this tier can be done more efficiently and timely than on the other tier, which would require accessing an entire 2MB of data to access the 512B in the block of interest.
  • the storage processor operates to move data between tiers of data stores with different characteristics and to move the data so the characteristics of the data reflect the characteristics of the data store. For example, if a particular block of data is being frequently written to the primary data storage system, the storage processor operates, if necessary, to move the data to a store tier that has higher performance characteristics that reflect the frequency with which the block of data is being written.
  • the primary data storage system maintains statistics that characterize units of data and uses these statistics in determining an appropriate tier for the data.
  • the storage processor uses units of data of various sizes in endeavoring to match the characteristics of the data to the characteristics of one of a number of tiers of data stores, the stores having different characteristics from one another.
  • FIG. 1 illustrates an embodiment of a networked computer system that includes an embodiment of a primary storage system
  • FIG. 2 is a block diagram of the management stack that processes administrator related communications, an I/O stack that processes communications relating to data storage, and fail-over stack that facilitates the transfer of responsibility for a volume between storage processors associated with the embodiment of the primary storage system shown in FIG. 1;
  • FIG. 2A illustrates an example of a statistics database that receives data from various elements of the primary data storage system and provides data to various elements of the system that, in many instances, use the data in performing a data storage related operation;
  • FIG. 3 illustrates an iSCSI encapsulation packet and an input/out block (IOB) derived from the packet;
  • IOB input/out block
  • FIG. 3 A illustrates the QoS attributes identified in FIG. 3;
  • FIG. 4 illustrates an example of a volume ownership table
  • FIG. 5 illustrates an example of a layer map and a volume information table
  • FIG. 6 illustrates an example of the operation of the QoS filter of the I O stack shown in FIG. 2 for a primary data storage system that services three initiators, each having a different criticality and different performance goals;
  • FIG. 7 illustrates an example of a journal and related journal table
  • FIG. 8 illustrates an example of a layer store table.
  • the networked computer system includes a user level 22, an initiator level 24, a first switch level 26 that facilitates communication between the user level 22 and the initiator level 24, a primary data storage level 28, a second switch level 30 that facilitates communications between the initiator level 24 and the primary data storage level 28, and a secondary data storage level 32.
  • system 20 includes a user level 22, an initiator level 24, a first switch level 26 that facilitates communication between the user level 22 and the initiator level 24, a primary data storage level 28, a second switch level 30 that facilitates communications between the initiator level 24 and the primary data storage level 28, and a secondary data storage level 32.
  • the user level 22 includes at least one user computer that is capable of being used in a manner that interacts with the primary data storage level 28.
  • a user computer is capable of requesting that: (a) data associated with the user computer be sent to the primary data storage level 28 for storage and (b) data stored in the primary data storage level 28 be retrieved and provided to the user computer.
  • At least one user computer associated with the user level is a storage administrator computer 34 that provides a storage administrator or system administrator with the ability to define the manner in which the data storage provided by the primary data storage level 28 is utilized.
  • the user level 22 typically includes a plurality of user computers with at least one of the plurality of user computers being associated with a storage administrator and the other user computers being associated with other entities.
  • the user level 22 includes user computers 36A-36C respectively associated with a customer support department, an accounting department, and an engineering department.
  • the initiator level 24 includes at least one initiator that operates to translate a request from a user computer into one or more block command packets.
  • a request from a user computer is in the form of a request packet that conforms to a packet protocol such as TCP, IP, Web, DB, and FileShare.
  • a block command packet conforms to a block protocol that includes block commands for data storage devices that operate on one or more blocks of data. Examples of block protocols are the Internet Small Computer System Interface protocol (iSCSI), the Fiber Channel protocol (FC), TCP, and IP.
  • block commands include: (a) a block write command that directs a data storage device to write one or more blocks of data to storage media associated with the device and (b) a block read command that directs a data storage device to read one or more blocks of data from a storage media associated with the device.
  • a block of data is a fixed and predetermined number of contiguous bytes of data that is or will be resident on some kind of storage media. Typical block sizes are 512, 1024, 2048, and 4096 bytes. For example, a request from a user computer to read a large file of data resident at the primary data storage level 28 is likely to be translated by an initiator into multiple block command packets that each relate to one or more blocks of data that is/are part of the requested file.
  • the initiator also operates to translate a block result packet, a packet that is received by the initiator and provides the result or a portion of the result of the execution of a block command associated with a block command packet, into a reply to request packet.
  • the initiator provides the reply to the request packet to the appropriate user computer.
  • the initiator level 24 commonly includes a plurality of initiators with each of the initiators capable of: (a) processing request packets from each of the user computers to generate block command packets and (b) processing block result packets to produce reply to request packets that are provided to the appropriate user computers.
  • the initiator level includes initiators 38A-38C.
  • An initiator can be comprised of a cluster of two or more computers that each endeavors to process a request from a user computer and that provide redundancy in the event that one or more of the computers fail.
  • an initiator that is designated to process high priority or critical requests is comprised of multiple computers, thereby providing redundancy should any one of the computers fail.
  • the first switch level 26 provides the ability for one or more user computers at the user level 22 to communicate with one or more initiators at the initiator level 24. More specifically, the first switch level 26 operates so as to receive a request packet from a user computer, process the request packet to determine which initiator should receive the request packet, and routes the request packet to the appropriate initiator. Conversely, the first switch level also operates to receive a reply to request packet from the initiator level 24, process the reply to request packet to determine which user computer should receive the reply to request packet, and routes the reply to request packet to the appropriate user computer.
  • the first switch level 26 can include a single switch that connects one or more user computers to one or more initiators or multiple switches that each connects one or more user computers to one or more initiators.
  • the first switch level 26 includes a switch 40 that is capable of establishing communication paths between the user computers 34 and 36A-36C and the initiators 38A-38C.
  • the primary data storage level 28 (or primary data storage system 28) operates to receive a block command packet from an initiator, attempt to execute the block command contained in the block command packet, produce a block result packet that contains the result of the attempted execution or execution of the block command, and provide the block result packet to the initiator that sent the related block command packet to the primary data storage system 28.
  • Typical block commands include a write command and a read command.
  • the primary data storage system 28 attempts to write one or more blocks of data to a data store (sometimes referred to simply as a "store") associated with the primary data storage system 28.
  • a read command the primary data storage system 28 attempts to read one or more blocks of data from a data store associated with the primary data storage system 28 and provide the read data to the initiator.
  • the primary data storage system 28 includes at least one storage processor and at least one data store.
  • the primary data storage system 28 also includes at least one switch when the at least one storage processor and the at least one data store associated with the at least one storage processor will accommodate two or more independent communication paths between the at least one storage processor and the at least one data store.
  • a storage processor includes an application memory and a processor for executing code resident in the application memory to process a block command packet.
  • the processor and the application memory are embodied in a SuperMicro Superserver 6036ST.
  • a data store is (a) a single data storage device or element or (b) a combination of data storage devices or elements.
  • Examples of a single data storage element that can each be a data store include a CPU bus memory, a disk drive with a magnetic/optical disk, a solid state drive, and a tape drive with a tape.
  • An example of a combination of data storage devices or elements that are configured to operate as a single data store is a group of disk drives configured as a Redundant Array of Independent Drives or RAID.
  • a data store can be characterized by the attributes of path redundancy, data redundancy, and persistence.
  • the path redundancy attribute is a measure of the number of redundant and independent paths that are available for writing data to and/or reading data from a data store. As such, the value of the path redundancy attribute is the number of independent paths (i.e., the independent I/O ports associated with the data store) less one. The value of the path redundancy attribute is one or greater when there are at least two independent paths available for writing data to and/or reading data from the data store. If there is only one independent path available for writing data to and/or reading from a data store, the path redundancy is zero.
  • the data redundancy attribute is a measure of the number of failures of elements in a data store that can be tolerated without data loss. As such, the value of the data redundancy attribute is the number of elements in the data store less the number of elements that can fail before there is data loss. For example, if a data store is comprised of two disk drives (elements) with the data on one disk drive mirroring the data on the other disk drive, the value of the data redundancy attribute is one because the failure of one disk drive means that the data can still be recovered but the failure of both disk drives would mean that there would be data loss.
  • the value of the data redundancy attribute of a RAID-6 data store comprised of six disk drives (elements) is two because the two of the disk drives (elements) can fail and the data can still be recovered but the failure of three or more disk drives (elements) would preclude the recovery of the data.
  • the switch facilitates communications between each of the storage processors or a subset of all of the storage processors associated with the primary data storage level 28 and each port of all of the data stores associated with the primary data storage system 28 or a subset thereof.
  • redundancy that allows the primary data storage system 28 to continue operation in the event of a predetermined level of failure of a storage processor, an element of a data store, and or a switch is desired.
  • This redundancy refers to path redundancy in which there are at least two separate and independent paths extending at least part of the way between an I/O interface of the primary data storage system 28, the interface that initially receives a block command packet from an initiator and from which a block result packet is transmitted to an initiator, and a data store.
  • partial path redundancy is provided by rendering at least two of the network cards 44A-44D with the same initiator. If one of the at least two Ethernet cards fails, the other operative Ethernet card(s) provide(s) path redundancy for the initiator.
  • Partial path redundancy is provided by the two storage processors 46A, 46B. If one of the first and second storage processors 46A, 46B fails, the other storage processor can be utilized to provide the path redundancy between the I/O interface 42 and a data store. In this regard, the non-failing storage processor may use one or both of the switches 50A, 50B.
  • the storage processor 46A can utilize both the first and second switches 50A, 50B to complete a communication path between the Ethernet card 44A and the first data store system 48A, i.e., the storage processor 46B utilizes the first and second switches 50A, 50B to communicate with both the Ethernet card 44A and the first data store system 48A.
  • Partial path redundancy is provided by the first and second switches 50A, 50B. If one of the first and second switches 50A, 50B fails, the other switch can be utilized to provide the necessary path redundancy. This path redundancy is dependent upon the non- failing switch having: (a) access to a portion of the data store that provides data redundancy relative to the portion of the data store that is no longer accessible due to the failure of the other switch and (b) access to an Ethernet card that can be addressed by the same initiator as the Ethernet card(s) that is/are no longer available due to the failure of the other switch.
  • Ethernet cards 44A and 44C are each addressable by the same initiator, the data store systems 48A and 48B each include an element that together define a data store in which one element mirrors the other element, and switch 50A fails, the switch 50B can be utilized to establish the necessary communication between the Ethernet card 44C and the element in data store system 48B.
  • the first data store system 48A is comprised of: (a) a first data store that is a first CPU bus memory 52A (sometimes referred to as memory store 52A) and is relatively fast but with relatively low capacity and no redundancy, (b) a second data store that is a first solid state disk or drive (SSD) 54A with less speed but greater capacity relative to the first CPU bus memory 52A and no redundancy, and (c) a third data store in the form of a first RAID disk array 56A with less speed and greater capacity than the first solid state disk 54A and redundancy.
  • a first data store that is a first CPU bus memory 52A (sometimes referred to as memory store 52A) and is relatively fast but with relatively low capacity and no redundancy
  • SSD solid state disk or drive
  • CPU bus memory is memory that is accessible to a processor of a storage processor via the processor's address bus, available for use by the processor, useable by the processor in processing a block command packet, and does not contain any portion of the application program that is executed or could be executed in the processing of a block command packet.
  • the processor accesses the first SSD 54A and the first RAID disk array 56A via an expansion bus (e.g., PCIe).
  • an expansion bus e.g., PCIe
  • stores having similar characteristics are typically configured within a primary data storage system so as to constitute a tier.
  • the first data store system 48A can be comprised of other combinations of partial data stores and/or data stores.
  • the first data store system 48A could include a first disk drive and the second data store system 48B could include a second disk drive, the first and second disk drives together forming a data store in which the first and second disk drives mirror one another to provide data redundancy.
  • the second data store system 48B includes data stores in the forms of a second CPU bus memory 52B (sometimes referred to as memory store 52B), a second SSD 54B, a second RAID disk array 56B. It should be appreciated that the second data store system 48B can also include other combinations of data stores and partial data stores.
  • the switch that is used to establish connections between the processor of a storage processor and the data store system is comprised of a type A switch that establishes connections with the non-CPU bus data storage and a type B switch that establishes connections with the CPU bus memory.
  • the first and second switches 50A, 50B respectively include type B switches 60A, 60B that respectively allow the processors of the storage processors 46A, 46B to establish communication paths with the CPU bus memories 52A, 52B.
  • a type B switch is comprised of the hardware, software, and/or firmware associated with a storage processor that allow the processor to access the memory locations on the CPU memory bus associated with the CPU bus memory.
  • the first and second switches 50A, 50B respectively include type A switches 58A, 58B that respectively allow the processors of the storage processors 46A, 46B to establish communication paths with the non-CPU bus data stores.
  • a type A switch is comprised of the hardware, software, and/or firmware associated with an expansion bus that allows the processor to access the data on the non-CPU bus data storages.
  • the second switch level 30 provides the ability for each of the initiators associated with the initiator level 24 to communicate with at least one network card associated with the primary data storage system 28, the at least one network card being associated with at least one storage processor of the primary data storage system 28. More specifically, the second switch level 30 operates to receive a block command packet from an initiator and process the block command packet so as to route the packet to the address that is associated with a particular network card. Conversely, the second switch level 30 also operates to receive a block result packet from the primary data storage system 28 and process the block result packet so as to route the packet to the appropriate initiator.
  • the secondary data storage level 32 also includes a cloud storage provider 64 that is accessible to the primary data storage system 28.
  • the cloud storage provider 64 can be a part of a data store, part of multiple data stores, a data store, multiple data stores, or combinations thereof that is respectively accessible to the storage processors 46A, 46B via network cards 66A, 66B (which are Ethernet cards in the illustrated embodiment) and the type A switches 58A, 58B respectively associated with switches 50A, 50B.
  • the system administrator computer 34 respectively communicates with the storage processors 46A, 46B via network cards 68A, 68B and switches 50A, 50B.
  • the administrator computer 34 can also communicate with the storage processors 46A, 46B via one or more paths that include the first switch level 26, the initiator level 24, and the second switch level 30.
  • the administrator command packets are processed using a management stack 100.
  • the management stack 100 is embodied in software that is executed by the storage processor.
  • the management stack 100 operates to receive an administrator command packet that relates to the primary data storage system 28, processes the administrator command packet, and provides a reply packet, if appropriate.
  • the receiving, processing, and replying of an administrator command packet by the management stack 100 involves interaction with other software elements and hardware elements within the primary data storage system 28.
  • the software elements with which the management stack interacts are: an IO stack and, if there is another storage processor, a fail-over manager and a second management stack.
  • An example of a hardware element that interacts with the management stack 100 is a network card.
  • the management stack 100 operates to conduct communications with any other storage processors at the primary data storage system 28.
  • Examples of administrator command packets include packets that relate to the creation of a LUN/volume within the primary data storage system 28, the assignment of Quality-of- Service (QoS) goals for a LUN/volume, the association of a LUN/volume with an initiator, the configuration of a network card (i.e., the assigning of an address to the Ethernet card so that the card is available to one or more initiators), requesting of data/information on the operation of a LUN/volume, the destruction of a LUN, and maintenance operations.
  • QoS Quality-of- Service
  • the management stack 100 conducts communications with the 10 stack 102 that relate to a volume(s) for which the 10 stack 102 is responsible.
  • communications with the 10 stack 102 are communications that involve the creation of a volume, the assignment of QoS goals to a volume, the association of a volume with an initiator, the configuration of an network card, the acquisition of data/information relating to a volume or volumes for which the 10 stack 102 is responsible, and the destruction of a volume.
  • the Ethernet hardware driver 108 controls an Ethernet card so as to produce the electrical signals needed to receive a message, such as an administrator command packet, and transmit a message, such as reply packet.
  • the TCP/IP protocol processor 110 at the TCP level manages the reassembly (if needed) of two or more packets received by an Ethernet card into the original message (e.g., an administrator command packet) and the disassembly (if needed) of a message into two or more packets for transmission (e.g., a reply to an administrator command).
  • the TCP/IP protocol processor 1 10 at the IP level assures the addressing of packets associated with a message. With respect to received packets, the IP level confirms that each of the received packets does, in fact, belong to the IP address associated with the Ethernet card. With respect to packets that are to be transmitted, the IP level assures that the each packet is appropriately addressed so that the packet gets to the desired destination. With respect to a received message, the TCP level also recognizes the packet as requiring further routing through the management stack 100, i.e., to the Web protocol processor 112 or Telnet protocol processor 1 14. The TCP/IP protocol processor 1 10 also performs other processing in accordance with the protocols, e.g., ordering packets, checksum etc.
  • the file that is the subject of the file call can result in the management server 120 communicating with the IO stack 102, the fail-over manager 104, the management database 122, and/or another storage processor.
  • the management server 120 will communicate with the IO stack 102, the fail-over manager 104, the management database 122, and other storage processors.
  • the management server 120 will communicate with the IO stack 102 to obtain the necessary statistics on the particular volume.
  • the management server 120 in addition to processing administrator command packets that propagate down the management stack, also processes commands or requests for information from management servers associated with other storage processors. For instance, a "foreign" management server that is associated with a different storage processor than the management server 120 may have processed an administrator command packet setting forth a new administrator id/password. The foreign management server would update its management database and forward a command to the management server 120 to update the management database 122 with the new administrator id/password.
  • the management database 122 has three portions: (a) a local object portion to which only the management server 120 can read/write, (b) a shared object portion to which the management server 120 can read/write but can only be read by another management server, and (c) a shared object to which the management server 120 can read/write and to which another management server can read/write.
  • An example of a shared object to which the management server 120 can read/write but that can only be read by another management server is information that is specific to the storage processor with which the management server 120 is associated, e.g., CPU usage or CPU temperature.
  • An example of a shared object to which both the management server 120 and another management server can read/write is an administrator id/password.
  • FIG. 2 illustrates the IO stack 102, i.e., a group of processes that are executed by each storage processor associated with the primary storage level 28 in processing a block command packet relating to a particular block of data or multiple blocks of contiguous data.
  • the FC protocol processor 152 (a) manages the reassembly (if needed) of two or more packets received by a Fibre Channel card into the original block command packet and the disassembly (if needed) of a result packet into two or more packets for transmission, and (b) assures the addressing of packets associated with a received block command packet and associated with a reply packet.
  • FCoE Fibre Channel over Ethernet
  • a SCSI block command can also be conveyed over an Ethernet and according to Fibre Channel protocols.
  • the SCSI block command is embedded in a block command packet that conforms to the Ethernet and FC protocol.
  • the network processors 130 include the Ethernet hardware driver 108 and the FC protocol processor 152.
  • the primary data storage system 28 operates to process block commands, i.e., commands that relate to the reading of a block data from or writing of a block data to a storage medium.
  • the primary data storage system 28 can be adapted to operate with block commands other that SCSI commands.
  • the primary data storage system 28 can be adapted to operate on block commands that are conveyed over a network according to protocols other than Ethernet, TCP/IP or FC.
  • the filter stack 132 is comprised of a target driver filter 160, a group of foreground filters 162, and a group of background filters 164. Associated with the filter stack 132 are a filter manager 166 and a statistics database 168. Operations that involve executing or attempting to execute a SCSI block command flow "down" the stack, i.e. in the direction going from the target driver filter 160 and toward the group of background filters 164. In contrast, operations that involve generating or providing the result of the execution or attempted execution of a SCSI block command flow "up" the stack. Consequently, a filter involved in executing or attempting to execute a SCSI block command may also be involved in generating or providing the result of the execution or attempted execution of the SCSI block command.
  • the group of foreground filters 162 process a command IOB to: (a) cause whatever write/read related operation is required of a block command to occur and (b) cause one or more tasks needed to accomplish the read/write operation to occur in a fashion that endeavors to meet QoS goals.
  • the foreground filters 162 also process a result IOB as needed and provide the result IOB to the target driver filter 160.
  • the filter manager 166 operates to create (associate) the filter stack 132 with a volume (an identifiable unit of data storage), destroy (disassociate) a volume from the filter stack 132, and cooperates with the fail-over manager 104 and/or management server 120 to implement various volume related functions (e.g., using the management server 120 to inform "foreign" storage processors of the creation of a new volume).
  • the statistics database 168 receives statistical data relating to a volume from one or more filters in the filter stack 132, stores the statistical data, consolidates statistical data based upon data provided by a filter, stores calculated statistical data, and provides the stored statistical data to one or more filters in the filter stack 132 and to the management server 120.
  • the filter manager 166 operates to create (associate) the filter stack 132 with a volume (an identifiable unit of data storage), destroy (disassociate) a volume from the filter stack 132, and cooperates with the fail-over manager 104 and/or management server 120 to implement various volume related functions (e.g., using the management server 120 to inform "foreign" storage processors of the creation of a new volume).
  • the filter manager 166 receives a message from the Management Server 120 instructing filter manager 166 to create a new volume with a specific filter stack configuration. The filter manager 166 instantiates the filters and places them in the correct hierarchy based on the storage administrator request.
  • a portion of a statistics database 258 is described to facilitate the understanding of the use of the database 168 by various filters.
  • a portion of the database relates to hardware.
  • the portion that relates to hardware includes statistics relating to a CPU, a Solid-State Disk (SSD), and an Ethernet card.
  • a portion of the example of a portion of the statistics database 258 relates to volume related data.
  • the portion that relates to volume data includes statistics directed to three different criticalities, a volume, and an initiator.
  • statistic relating to throughput, queue depth, latency, and use count are provided.
  • the use count with the "second" resolution corresponds to IOPS.
  • the use count with respect to resolutions of greater duration are IOPS scaled to the resolutions of the greater duration.
  • statistics are provided in terms of both reads and writes.
  • the example of a portion of a statistics data includes current statistical data and historical statistical data.
  • the current statistical data has a resolution of "second.”
  • the historical statistical data has resolutions great than "second” and include resolutions of "minute", "hour”, and "day”.
  • Target Driver Filter The operation of the target driver filter 160 is described with respect to the processing of a type of block command packet, known as an iSCSI encapsulation packet 180 (sometimes referred to as "command packet") that includes a SCSI command, to generate an IOB 182.
  • the command packet 180 is a packet that encapsulates a SCSI block command and other information, is received at one of the Ethernet cards 44A-44D, and processed by the Ethernet hardware driver 108, TCP/IP protocol processor 110, and iSCSI protocol processor 140 prior to being provided to the target driver filter 160.
  • the target driver filter 160 can be adapted to operate with block commands other than SCSI block commands, networks other than the Ethernet, and network protocols other than TCP/IP.
  • the IOB 182 is a data structure that stores data/information associated with the processing of the SCSI block command. More specifically, the IOB 182 provides multiple fields for holding data/information relating to the processing of the SCSI block command.
  • the target driver filter 160 builds the IOB 182 and populates certain fields of the IOB with data/information from the command packet 180.
  • the IOB 182 is then provided to each of the other filters in the filter stack 132 that is involved in the executing or attempting to execute the SCSI command (i.e., going down the stack). Each of these other filters can, if needed, read data/information from one or more fields in the IOB 182 and, if needed, write data/information to one or more fields in the IOB 182.
  • the IOB 182 is then provided to each of the filters in the filter stack 132 that is involved in providing the result of the of the processing of the SCSI command (i.e., going up the stack). Ultimately, the IOB 182 is provided to the target driver filter 160 which uses the IOB 182 to create an iSCSI encapsulation packet that includes the result of the processing of the SCSI command, i.e., a result packet. The result packet is then provided to the network processors 130 for additional processing and transmission of the results packet towards the initiator that originated the command packet.
  • the IOB 182 is comprised of an Initiator ID field 220, a VolID field 222, a PageMode field 224, an LBA/PageNum field 226, a SectorCount/PageOffset field 228, a Command field 230, an ErrorCode 232 field, an ErrorOffset field 234, a NumberOfDataSegments field 236, DataSegmentVector field 238, a DataCRCVector field 240, a Layerld field 242, a QoS attributes field 244, a StorelD field 246, a StoreLBA field 248, an In Time Stamp field 250, an Issuer stack field 252, and an XtraContext field 254.
  • the QoS attributes field 244 is comprised of a criticality field 260A, AllowedStores field 260B, AllowedLatency 260C, Projectedlmpact 260D, and ImpactArray 260E.
  • the Impact Array 260E includes impacts for each of the physical components of the primary data storage system (e.g., CPU, memory, SAS, SSD, and Ethernet) and the software components (e.g., volume, criticality, and initiator).It should be appreciated that the AllowedLatency 260C and the InTimeStamp 250 are used in a "headroom" evaluation (i.e., an evaluation as to the amount of time available to perform an operation) in such a way that as filters higher in the stack consume time operating on an IOB, the filters lower in the stack have less "headroom" to operate on the IOB.
  • headroom i.e., an evaluation as to the amount of time available to perform an operation
  • the target driver filter 160 builds the IOB 182 and populates certain fields of the IOB 182 with values from or derived from the command packet 180. It should be appreciated that a value associated with a field is sometimes referred to simply by the field name.
  • the target driver filter 160 uses data/information in the LU field 206 of the command packet 180 to derive a value for the VolID field 222 of the IOB 182, i.e., the volume within the primary data storage system 28 to which the SCSI block command relates.
  • the value in the VolID field 220 reflects the priority (e.g., mission critical, business critical, non-critical) that the administrator has associated with the data blocks that are associated with volume.
  • the target driver filter 160 sets the value of the PageMode field 224 to "off to indicate that the IOB 182 initially relates to a block or blocks of data within a volume and not to a block or blocks of data within a page, a larger unit of memory than a block.
  • the "off value in the PageMode field 224 also indicates that the values established or to be established in the LBA/PageNum field 226 and SectorCount/PageOffset field 228 are LBA and SectorCount values and not PageNum and PageOffset values.
  • the target driver filter 160 uses data/information in the SCSI Command Data Block field 208 to populate the command field 230 with the SCSI command (e.g., a block read command or a block write command), the LBA/PageNum field 226 with the address of the first logical block address within the volume to which the SCSI command relates, and the SectorCount/PageOffset field 228 with the number of sectors (or blocks) beginning at the specified LBA to which the SCSI command relates.
  • a block read command is referred to as a read block command.
  • a block write command is referred to as a write block command.
  • the ErrorCode field 232 holds an error code value that is subsequently established by a filter in the filter stack 132 and indicative of a type of error encountered in the processing of the SCSI command or in the returning of the result of the processing of the SCSI command.
  • the ErrorOffset 234 field holds an offset value that further defines the type of error identified in the ErrorCode field 232.
  • the target driver filter 160 uses the data segment field 198 to establish values in the NumberOfDataSegments field 236 and the DataSegmentVector field 238.
  • the target driver filter 160 places the data (sometimes referred to as "write data") in the Data Segment field 198 into memory (e.g., memory store 52A or 52B).
  • write data the data in the Data Segment field 198 into memory
  • the data from the Data Segment field 198 may be broken into two or more non-contiguous segments.
  • the target driver filter 160 places the number of data segments that are established in memory in the NumberOfDataSegments field 236 and the address and length of each of the segments established in memory in the DataSegmentVector field 238. If there is more than one segment established in memory, the target driver filter 160 calculates a cyclic redundancy check (CRC) or possibly another form of hash for each of the segments and places each of the CRC values in the DataCRCVector field 240. If there is only one segment established in memory (i.e., all of the data in the Data Segment field 198 was copied into a single segment in memory), the target driver filter 160 copies the value that is in the Data CRC field 210 to the DataCRCVector field 240. It should be appreciated that a data verification techniques other that CRC can be employed in place of CRC.
  • CRC cyclic redundancy check
  • the target driver filter 160 calculates a CRC on the data in the Data Segment 198 and compares the calculated CRC to the CRC value (if present) in the Data CRC field 210. If there is a difference between the calculated CRC and the CRC in the field 210, then the data in the Data Segment 198 has somehow been corrupted. In this case, the processing of the SCSI command is aborted and the target driver filter 160 prepares a result packet indicating that the command failed to execute. The result packet is passed on to the network processors 130 for processing and transmission to the initiator.
  • the IssuerStack field 252 is used by a filter in the filter stack 132 that is operating on a command IOB (i.e., when the flow of the IOB is down the filter stack 132) to indicate that the filter needs to do additional processing when the result IOB is propagating up the stack (i.e., when a result of the execution of the SCSI command or failure to execute the SCSI is being prepared).
  • the XtraContextStack field 254 is a field that a filter can use to store additional context information when the filter has indicated in the IssuerStack field 252 that the filter needs to do additional processing when the IOB is propagating up the stack.
  • the IssuerStack field 252 has a stack structure in which each filter that needs to do additional processing "pushes" down an indication of the need to do additional processing onto the "stack.” As a result IOB propagates up the stack, a filter that "pushed" down an indication of a need to do additional processing "pops" off or removes the indication from the IssuerStack field 252 after the additional processing of the IOB is completed by the filter.
  • the XtraContext Stack field 254 also has a push/pop structure that functions in a substantially similar way to the IssuerStack field 252.
  • [Para 113]Later when a result IOB 182 is propagating up the filter stack 132 and reaches the target driver filter 160, the current time is obtained, the "In" time stored in the In Time Stamp field 250 is obtained, and the total latency associated with the processing of the IOB is calculated, i.e., the elapsed time between when the "In" time value was obtained by the target driver filter 160 and the when the current time was obtained.
  • the target driver filter 160 updates initiator and volume tables in the statistics database 168 with the total latency value. It should be appreciated that other tables or statistics in the statistics database 168 may also be updated. Additionally, the target driver 160 builds the result packet and provides the result packet to the network processors 130 for further processing and communication to the initiator.
  • An initiator can send a command packet to the primary data storage system 28 that relates to a volume for which the storage processor that initially starts processing the IOB relating to the command packet is not responsible.
  • the I/O forward filter 270 operates to identify this situation and forward the IOB to the storage processor that is responsible for the volume.
  • the filter manager 166 associated with the storage processor creates the volume and updates a volume ownership table to indicate that the particular storage processor and no other storage processor in the primary data storage system 28 is responsible for the volume.
  • a volume ownership table 286 is illustrated.
  • the filter manager 166 indicates to the fail-over manager 104 that the volume ownership table has changed.
  • the fail-over manager 104 communicates that there has been a change in the volume ownership table to the fail-over manager associated with each of the other storage processors in the primary data storage system 28.
  • volume ownership table There are a number of other situations that cause a change in the volume ownership table and the change to be communicated to the other fail-over managers. For instance, the destruction of a volume causes such a change in a volume ownership table.
  • Another situation that causes a change in the volume ownership table is a fail-over, i.e., a situation in which the storage processor that is responsible for a volume cannot adequately service the volume and responsibility for the volume is transferred to another storage processor.
  • the volume ownership table identifies the volume(s) for which each storage processor in the primary data storage system 28 is responsible.
  • the I/O forward filter 270 forwards the IOB to the I/O forward filter associated with the "foreign" storage processor that the volume ownership table indicates is the "owner" storage processor of the volume.
  • the forwarding of the IOB involves the use of the switches 50A, 50B.
  • the result IOB is forwarded back to the I/O forward filter 270 of the "home" storage processor.
  • the "home" storage processor passes the result back up the stack so that the result can be placed in a result packet and sent to the originating initiator.
  • the creation of layer 3 in response to the snapshot command also created a snapshot of volume A that is reflected in layers 0, 1, 2 as of the time layer 3 was created.
  • the creation of layer 4 caused the volume information table 292 to be updated to show layer 4 as being the newest layer associated with volume B and to reflect a specified LBA offset of zero.
  • the creation of layer 4 also created a snapshot of volume B that is reflected in layers 1 and 0, with layer 1 being shared with volume A.
  • the creation of layer 5 caused the volume information table 292 to be updated to indicate that layer 5 is the newest layer associated with volume C and to show a specified LBA offset of zero.
  • the layer map filter 272 receives the IOB provided by the I/O forward filter 270 and processes the IOB to determine a layer id (LID) and a layer logical block address (LLBA) for the related SCSI command. More specifically, the layer map filter 272 uses the volume id specified in the VolID field 222 to index into the current volume information table 292 to determine the newest LID associated with the volume and LBA offset associated with the volume. The layer map filter 272 populates the LayerlD field 242 with the LID retrieved from the volume information table.
  • LID layer id
  • LLBA layer logical block address
  • the layer map filter 272 revises the LBA in the LBA/PageNum field 226 to reflect the LLBA, which is the current LBA value plus/minus the retrieved offset value.
  • the layer map filter 272 uses the LID and LBA to index into a layer-store table (e.g., FIG. 8) and retrieve the StorelD and StoreLBA values to populate the Storeld field 246 and StoreLBA field 248 of the IOB.
  • the quality-of-service (QoS) filter 274 generally provides predictable data storage performance to one or more initiators that utilize a shared data storage system (i.e., the primary data storage system) with multiple volumes.
  • the desired performance of a particular volume (criticality) is established by the administrator using the administrator computer 34 to communicate with the management stack 100.
  • the administrator uses the administrator computer 34 to create a volume
  • the administrator also uses the administrator computer 34 to associate a criticality with the volume.
  • the management stack 100 maintains a table/tables that identifies each of the initiators that the primary data storage system 28 will service and the criticality associated with each of the volumes that have been created.
  • Throughput is a measure of the number of bytes prepared for transfer (read/write) per unit of time within the filter stack 132 with respect to a volume.
  • IOPS is a measure of the number of IOBs processed within the filter stack 132 per unit of time with respect to a volume.
  • the specification of a criticality for a volume is embodied in a goal with respect to each of these attributes. It should be appreciated that a greater number, lesser number, and/or different attributes may be appropriate in certain situations. It should also be appreciated that two volumes with the same criticality can have the same or different quality of service or performance goals.
  • the performance of a data store in the primary data storage system 28 can also be characterized in terms of latency, throughput, and IOPS. Further, this "store performance" of a data store is or may be relevant to whether the performance goals with respect to a volume are being met. As such, the production of statistics relating to the "store performance" of data stores in the primary data storage system 28 are produced and available for use in assessing performance with respect to a volume. Further, other hardware and software in the primary data storage system 28 are also be characterized and monitored for use in assessing performance with respect to a volume.
  • the QoS filter 274 operates to sort IOBs that are associated with different volumes having different criticalities (i.e., different performance goals) so as to try to meet the goals of each volume. More specifically, the QoS filter 274 receives an IOB from the layer map filter 272 and processes the IOB to perform: (a) a first sort of the IOB according to the volume ID, i.e., according to the criticality associated with the volume, (b) a second sort of the IOB according to the projected impact of the processing of the IOB on the data storage system at the primary data storage system 28, the projected impact taking into account certain metrics/statistics relating to the operation of the primary data storage system 28, and (c) a third sort of the IOB into an IOB execution stack based upon the criticality associated with the volume identified in the IOB (first sort), the projected impact (second sort), past usage of the primary data storage system 28 as reflected in certain metrics/statistics, the current state of the primary data storage system 28 including the state of each
  • the QoS filter 274 places each IOB that is received from the layer map filter 272 into first-in-first-out input queue 300.
  • the QoS filter 274 processes each of the IOBs in the queue 300 in the order that the IOB was received in the queue 300. The following describes the further processing of the IOB 182 by the QoS filter 274.
  • each of the goal schedulers 304A-304C processes an IOB received from the group schedule 302 to assess the IOB as to the projected impact of the execution of the SCSI command.
  • each IOB is assessed as to whether execution of the SCSI command is likely to primarily affect latency, throughput, or IOPS.
  • the assessment takes into account metrics/statistics obtained from the statistics database 168. These metrics/statistics include volume related statistics.
  • statistics relating specifically to the volume with which the IOB is associated can be used.
  • statistics relating an initiator i.e., statistics relating to a number of volumes associated with a specific initiator
  • the statistics can include any number of factors, including throughput, queue depth, latency, and use count for these volume related statistics. However, currently it is believed that at least latency statistics are needed. Further, these factors can further include read and write related versions of each of throughput, queue depth, latency, and use count. Moreover, these factors can include current and historical statistics.
  • Current statistics being those statistics associated with the shortest period of time (or shortest resolution) and historical statistics being statistics associated with a greater period or periods of time relative to the shortest period of time. See, example of a portion of a statistics database 258.
  • the use of statistics relating to "criticality" and/or historical statistics facilitates the identification of imbalances and the like in the processing of IOB associated with volumes having the same criticality. For example, if the processing of IOBS associated with one volume has placed another volume with the same criticality increasingly behind its quality of service goals, the statistical data provides a basis for identifying this issue and taking action to bring the lagging volume back towards its quality of service goals.
  • the assessment results in the IOB being placed in one of a latency queue, throughput queue, and IOPS queue associated with the goal scheduler.
  • the goal scheduler also stores the result of the assessment in the IOB Projectedlmpact 260D field of the QoS Attributes 244.
  • the QoS filter 274 includes a shared hardware scheduler 312 that assesses the IOBs that are the next in line to be processed in each of the latency, throughput, and IOPS queues (the IOBs that are at the "bottom" of each of the queues) to determine which IOB will be placed in or merged into an FIFO execution queue 314, i.e., a queue that defines the order in which the IOBS received at the input queue 300 are to be executed.
  • the assessment of each of the IOBs takes into account the criticality and projected impact of the execution of the command associated with the IOB that is set forth in the QoS attributes field of each IOB and metrics/statistics obtained from the statistics database 168.
  • These statistics include hardware related statistics. For example, statistics relating the CPU, Ethernet cards, and stores (e.g., SSD) can be employed. These factors can include throughput, queue depth, latency, use count. Further, current and/or historical versions and/or read and/or write versions of these factors can be used. It should be appreciated that the comparison of the IOBs from the goal scheduler output queues to one another are comparisons of different volumes that have different criticalities and different quality of service goals (IOPs, throughput, and latency). For example, if the next selected IOB is throughput related the shared hardware scheduler 312 will use information in the statistics database 168 to determine a store that has available bandwidth to process the command and send the IOB down the stack "tagged" with that store as the destination.
  • IOPs criticality and different quality of service goals
  • the shared hardware scheduler 312 evaluates each of these IOBs to determine which one of IOBs 316A-3161 is the next to be placed in the execution queue 314. If, for example, the shared hardware controller 312 decided that IOB 316A was the next to be placed in the execution queue 314, the next evaluation by the shared hardware controller 312 would be with respect to IOBs 316B-3161 and IOB 316J, which has taken the place of IOB 316A at the head of the IOPS queue 31 OA. Before an IOB is placed in the execution queue 314, the related IOB is updated so as to "push" an indication onto the IssuerStack field 252 that the QoS filter 274 needs to do additional processing on the IOB when the IOB is propagating up the filter stack 132.
  • Fig. 6 shows a specific implementation of the QoS filter 274.
  • the QoS filter 274 is more generally characterized as producing a sum of weighted factor values for an IOB that indicate or signify the rank of the IOB relative to other IOBS being processed.
  • the factors can include the volume and hardware related throughput, queue depth, latency, use count, the noted current-historical-read-write versions thereof. The values for these factors are obtained from the IOB and the statistics database.
  • the weighted coefficients associated with each factor being dynamically adjustable to reflect the changing priorities with respect to the volumes and hardware due to what is typically a changing workload being placed on the system.
  • VolumeStats Stats . GetVolumeMetrics (lob .VolID)
  • InitiatorStats Stats . GetlnitiatorMetrics ( lob . InitiatorID)
  • CriticalityStats Stats . GetCriticalityMetrics (MyCriticality) /* Calculate a weighted sum related to */
  • IOPsConst [SourceVolumeSec] + VolumeStats . IOPs [ Sec ] *
  • IOPsConst [SourceVolumeMinute] + VolumeStats .
  • IOPsConst [SourcelnitiatorSec] + InitiatorStats .
  • IOPsConst [SourceCriticalitySec] + CriticalityStats .
  • LatencyConst [SourcelOB] + IoB . InTimeStamp *
  • LatencyMultiplier [ SourcelOB] + LatencyConst [SourceVolumeSec] + VolumeStats . Latency [ Sec ] *
  • LatencyConst [SourcelnitiatorSec] + InitiatorStats . Latency [ Sec ]
  • LatencyConst [ SourceCriticalitySec ] +
  • IOPsConst [ASource] IOPsConst [ASource ] *
  • IOPsMultiplier [ASource] IOPsMultiplier [ASource ] *
  • LatencyConst [ASource ] LatencyConst [ASource ] *
  • LatencyMultiplier [ASource ] LatencyMultiplier [ASource ] *
  • LoadCoefecientsArray (Coefecients, ACriticality, Almpact, Acomponent)
  • LoadAdj ustmentArray (Adj ustment, ACriticality, Almpact, Acomponent)
  • DominateFactor 0
  • WeightForlob Possiblelob .
  • ComponentStats Stats . GetMetrics (AComponent)
  • ACriticality [ACriticality] [AComponent] [AResolution] - Adj ustment [ACriticality] [AComponent] [AResolution];
  • the statistics filter 276 operates to collect certain initiator and volume related data/statistical information for each IOB passed to the statistics filter 276 from the QoS filter 274 when the IOB is going down the filter stack 132.
  • the statistics filter 276 processes the IOB 182 to obtain the initiator id from the InitiatorlD field 220, the volume id from the VolID field 222, the sector count from the SectorCount/PageOffset field 228, and the "In" time stamp value from the In Time Stamp field 250.
  • the statistics filter 276 also obtains the current time from the operating system.
  • the statistics filter 276 uses the value of the "In” Time Stamp and the current time to calculate the latency that the IOB has experienced between when the "In” Time Stamp value was established in the target driver filter 160 and when the current time is obtained by the statistics filter 276 (hereinafter referred to as "first latency").
  • the statistics filter 276 communicates with the statistics database 168 so as to: (a) update a table for the initiator that is maintained in the database to reflect that an IOB associated with the initiator will be processed that has the sector size obtained from the IOB and that the IOB has experienced the calculated first latency and (b) update a table for the volume that is maintained in the database to reflect that an IOB associated with the volume will be processed that has the sector size obtained from the IOB and that the IOB has experienced the calculated first latency.
  • the statistic filter 276 also pushes an indication onto the IssuerStack field 252 of the IOB 182 that the statistics filter 276 needs to do additional processing when the IOB is propagating up the filter stack 132. Further, the statistic filter 276 also pushes the current time onto the XtraContextStack field 254.
  • the statistics filter 276 when the IOB 182 is propagating up the filter stack 132 and reaches the statistics filter 276, the statistics filter 276 obtains the time from the XtraContextStack field 254 (which is no longer the current time), obtains the "new" current time, and calculates a second latency, i.e., the elapsed time between when the time value was obtained that was pushed onto the XtraContextStack field 254 and the IOB was propagating down the filter stack 132 and the when the "new" current time was obtained.
  • the statistics filter 276 updates the initiator and volume tables in the statistics database 168 with the second latency value. Further, the statistics filter 276 uses the values from the ImpactArray 260E to update the statistics database 168. When updating the database it may be necessary to update multiple rows of data, (e.g. when updating the CPU statistics it may be required to update the row for Second, Minute, Hour, and Day).
  • the pattern de-duplication filter 278 operates to preserve storage capacity and reduce turn around time to the initiator at the primary data storage system 28 by preventing a block(s) of identical data that are frequently written to the primary data storage system 28 from being written multiple times with each such writing of the block(s) of data consuming additional storage capacity and time. More specifically, the pattern de-duplication filter 278 operates to identify a block(s) of data that have a pattern which can be readily calculated. Characteristic of a pattern is that the values of each byte of data in a block can be calculated.
  • a pattern that can be "readily" calculated is a pattern that can be calculated or retrieved and the IOB completely processed (i.e., a result packet is prepared) within the latency associated with the volume. It should be appreciated that, for a given latency, the number of patterns that can be readily calculated increases with increasing processing speed.
  • the pattern de-duplication filter 278 makes a "headroom" calculation to determine if there is sufficient time available to perform the operations associated with pattern deduplication, which includes the time needed to identify a calculation engine that may be able to calculate a pattern associated with the write data and the time needed to determine if there is a match between the write data and the data produced by the selected calculation engine. In this regard, there needs to be sufficient time to conduct these operations within whatever time remains in the allowed latency 260C.
  • the pattern de-duplication filter 278 assesses data in the first block of data associated with each IOB having a SCSI write-related command to determine if a known calculable pattern of data is present. If all of the data in the first data block has a known calculable pattern, the pattern de-duplication filter 278 proceeds to assess the second and any additional blocks of data associated with the IOB. If all of the data in all of the blocks of data associated with the IOB have a known calculable pattern, there are two possibilities.
  • the pattern de- duplication filter 278 updates the command field 230 of the IOB so as to reflect that a de-dup write needs to be done and passes the IOB down the filter stack 132.
  • the pattern de-duplication filter 278 places a "success" code in the error code field 232 and causes the IOB to start propagating up the filter stack 132, thereby indicating that the SCSI write command of the IOB has been completed.
  • the pattern de-duplication filter 278 determines the pattern de- duplication is not possible and passes the IOB on to the dictionary de-duplication filter 280.
  • the pattern de-duplication filter 278 avoids doing so by making an initial comparison of two bytes in a block of data and using the result of the comparison for concluding that the data in the block: (a) potentially has one of the known calculable patterns or (b) does not possess one of the known calculable patterns.
  • This two byte comparison is a form of a "hash" calculation. It should be appreciated that methods other than the noted two byte comparison (a form of hash) can be applied (e.g. CRC or hash) as long as the methods can make the determination within the latency constraint, i.e., the allowed latency set forth in volume information table 292. If the comparison indicates that the data in the block potentially has one of the known calculable patterns, the pattern de- duplication filter 278 proceeds to assess the data in the block to determine whether the data in the block actually does have the identified, known calculable pattern.
  • the pattern de-duplication filter 278 utilizes the pattern calculator to calculate the value that a byte(s) of the pattern should have if present in the data block and compare each such value to the actual value associated with the byte(s) in the data block.
  • a calculator that is efficient, i.e., makes a determination of whether or not the pattern is present in the data more quickly rather than less quickly so as to make the determination within the latency constraint, i.e., the allowed latency set forth in volume information table 292.
  • the comparison is done in the fastest data store available, typically memory store 52A and 52B.
  • a relatively efficient calculator for determining if this wave pattern is present in a block would: (a) with respect to the potential first cycle of the wave pattern in the block, use the pattern calculator to calculate a first value for the wave pattern and compare that value to the two bytes in the data that should have the calculated value if a first cycle of the triangle wave is present in the block and (b) repeat this calculation and comparison to the values associated with different bytes in the data block until the presence of the first cycle of a triangle wave in the data is either confirmed or disaffirmed.
  • the pattern de-duplication filter 278 passes the IOB on to the dictionary de-duplication filter 280. If the presence of a first cycle of the triangle wave in the data is confirmed, the calculator proceeds to compare the data associated with the first cycle of the triangle wave to the data in the block that might be the second cycle of the triangle wave to either confirm or disaffirm the presence of the second cycle of the triangle wave. If the second cycle of the triangle wave is not present, the pattern de- duplication filter 278 passes the IOB on to the dictionary de-duplication filter 280.
  • the calculator proceeds to compare the data associated with the first and second cycles of the triangle wave to the data in the block that might be the third and fourth cycles of the triangle wave. This process of comparing groups of bytes that increase in number by a factor of two with each comparison continues until either the presence of the pattern in all of the blocks associated with IOB is confirmed or disaffirmed.
  • the pattern de-duplication filter 278 operates on an IOB having a SCSI read-related command to determine if the data at the identified volume id and LBA is data that has been previously de-duplicated in the processing of an IOB with a SCSI write-related command. More specifically, the pattern de-duplication filter 278 obtains the value in the StorelD field 246. If the value in the StorelD matches a StorelD assigned to a calculator engine (e.g., engine 320), the pattern de-duplication filter 278 concludes that the read-related command in the IOB relates to pattern data that has been de-duplicated.
  • a calculator engine e.g., engine 320
  • the de-duplication filter 278 obtains the value in the StoreLBA field 248 to identify the vector into the calculator for calculating the particular pattern and uses the calculator to create the block(s) of patterned data in the memory store (e.g., CPU bus memory 52A or CPU bus memory 52B), if the block(s) of patterned data do not already exist in the memory store.
  • the pattern de-duplication filter 278 then updates the value in the DataSegmentVector field to point to the address in the memory store (e.g., CPU bus memory 52A or 52B) that has the copy of the calculated pattern.
  • the pattern de-duplication filter 278 places a "success" code in the error field 232 and causes the IOB to start propagating up the filter stack 132, thereby indicating that the SCSI read-related command of the IOB has been completed. If the value in the StorelD does not match a StorelD assigned to a calculator engine, the IOB is passed down the filter stack 132 for further processing.
  • IdentifyingValueB] 234 main ( ) ⁇
  • Engineldx 0 ; Engineldx ⁇ 255; Engineldx ++ ⁇
  • EngineChoice DetermineEngineCandidate ( lob)
  • Buffer [bytenum] GenByteB ( lob . StoreLBA,
  • the dictionary de-duplication filter 280 operates to preserve storage capacity and reduce turn around time to the initiator at the primary data storage system 28 by preventing blocks of data associated with an IOB that constitute a page (a predefined number of contiguous blocks of data) that are commonly written to the primary data storage system 28 and do not have a readily calculable pattern from being written multiple times such that each writing of the page consumes additional storage capacity and time.
  • the dictionary de-duplication filter 280 has access to a dictionary table that is capable of holding a limited and predetermined number of entries.
  • Each non-null entry in the dictionary table relates to a page of data identified by an advanced de-duplication filter, one of the background filters 164, as being one of the most common pages of data being written to storage. More specifically, each non-null entry in the dictionary table for a "dictionary" page has StorelD and StoreLBA values for a copy of a "dictionary" page that is on a dictionary store 322.
  • the dictionary de-duplication filter 280 is one of the group of foreground filters and speed of execution is a priority in the foreground
  • the dictionary store 322 that holds the copy of the "dictionary" page is typically a high-speed store, like memory store 52A or memory store 52B.
  • the entry in the dictionary table also identifies a portion of data in the relevant "dictionary" page (e.g., the second 64- bytes of data in the page) that is unique relative to all of the other non-null entries in the dictionary table.
  • each entry has the first 64-bytes of a first "dictionary” page and another entry has the second 64-bytes of a second "dictionary” page
  • the use of the same identifying portion of data from each of the "dictionary" pages facilitates the assessment of whether the page associated with an IOB can be de-duplicated.
  • the data in the identifying portion relating to a single "dictionary" page is unique relative to all the other non-null entries in the dictionary table. Because the most commonly written pages can change over time and the dictionary table has a limited and predetermined number of entries, the advanced de-duplication filter can change the entries in the dictionary table. In this regard, a change to the table may require that a different identifying portion of the pages to which the entries in the table relate be used to preserve the uniqueness of each entry in the table.
  • the identifying portion of each of the dictionary pages that is unique is maintained by the advanced de-duplication filter and available to the dictionary de-duplication filter 280.
  • the advanced de-duplication filter also ensures that a copy of each of the common pages that is identified in dictionary table is in the dictionary store 322.
  • the dictionary de-duplication filter 280 makes a "headroom" calculation to determine if there is sufficient time available to perform the operations associated with dictionary deduplication, which includes the time needed to identify a dictionary entry that may correspond to the write data and the time needed to determine if there is a match between the write data and the data in the dictionary entry. In this regard, there needs to be sufficient time to conduct these operations within whatever time remains in the allowed latency 260C.
  • the dictionary de-duplication filter 280 determines if the write command relates to a page.
  • This determination is made by obtaining the sector count value in the SectorCount/PageOffset field 228 in the IOB. If the value is not equal to the number of blocks in a page, the dictionary de-duplication filter 280 passes the IOB on down the filter stack 132. If, however, the value is equal to the number of blocks in a page, the dictionary de-duplication filter 280 obtains the same portion of the page associated with the IOB that is associated with the identifying portion in each entry in the dictionary table and compares this portion of the page to each identifying portion in the dictionary table.
  • the dictionary de-duplication filter 280 passes the IOB on down the filter stack 132. If there is a match, then there is a possibility that the page associated with the IOB is a match with the "dictionary" page to which the entry in the dictionary table relates. To determine whether there is such a match, the dictionary de-duplication filter 280 compares the page associated with the IOB to the copy of the "dictionary" page that is located at the StorelD and StoreLBA of the dictionary store 322 set forth in the dictionary table.
  • the data associated with the write IOB and the dictionary page are both in memory store 52A or 52B, the fastest type of store in the illustrated system. As such, the comparison occurs more quickly than if the comparison was done in some other store in the system. If there is no match, the dictionary de-duplication filter 280 passes the IOB down the filter stack 132. If there is a match, there are two possibilities.
  • the dictionary de-duplication filter 280 operates on an IOB having a SCSI read-related command that need not relate to a page to determine if the data associated with the identified volume id and LBA is data that has been previously de-duplicated in the processing of an IOB with a SCSI write-related command relating to the same volume id and LBA. More specifically, the dictionary de- duplication filter 280 obtains the value in the StorelD field 246 and determines if the value is currently associated with the dictionary store 322.
  • the dictionary de-duplication filter 280 then updates the value in the DataSegmentVector field to point to the address in the memory store (e.g., memory store 52A or 52B) that has the copy of the dictionary page and, more specifically, to point the first block of the page that has the first block to which the SCSI read command relates. Further, the dictionary de-duplication filter 280 places a "success" code in the error field 232 and causes the IOB to start propagating up the filter stack 132, thereby indicating that the SCSI read- related command of the IOB has been completed. If the value in the StorelD field 246 is not currently associated with the dictionary store 322, the IOB is passed down the filter stack 132 for further processing.
  • the memory store e.g., memory store 52A or 52B
  • HitCount [DictionaryMax] 0, 0, 0, 0, 0 main ( ) ⁇
  • Bufldx 0 ; Bufldx ⁇ DictionaryMax ; Bufldx ++ ⁇
  • TestOffset 0 ; TestOffset ⁇ 512 ; TestOffset ++ ⁇
  • Bufldx 0 ; Bufldx ⁇ DictionaryMax ; Bufldx ++ ⁇
  • Iob.StoreLBA StoreLba [PossibleBuffer]
  • Bufldx 0 ; Bufldx ⁇ DictionaryActive ; Bufldx ++ ⁇
  • Bufldx 0 ; Bufldx ⁇ DictionaryActive ; Bufldx ++ ⁇
  • Bufldx 0 ; Bufldx ⁇ (DictionaryActive - 1) ; Bufldx ++ ⁇ if (HitCount [Bufldx] ⁇ HitCount [Bufldx + 1]) ⁇
  • the I/O journal filter 282 operates with respect to IOBs in the execution queue 314 that have SCSI write-related commands (de-dup write and write) that have not been fully addressed by an intervening filter to move the actual data that is associated with the IOBs and currently resident in a non-redundant and/or non- persistent data store or other information that allows the data to be reproduced to a redundant and persistent data store (i.e., a journal store). Further, because the I/O journal filter is part of the foreground filters 162, the I/O journal filter 282 endeavors to do so in a timely fashion.
  • the I/O journal filter 282 Because the actual data associated with an IOB or other information that allows the actual data associated with the IOB to be reproduced is moved to a redundant and persistent data store, the I/O journal filter 282 also causes each such IOB to begin propagating up the filter stack 132, thereby acknowledging completion of the write-related command. There are two characteristics of the I/O journal filter 282 that each contribute to the timely processing. The first characteristic is that each write to the redundant and persistent store is the writing of a page, which is comprised of a large number of blocks. As such, for a given number of data blocks, the writing of pages requires fewer writes relative to an approach in which there is a separate write operation for each block.
  • the second characteristic is that the writes are done to locations in the redundant and persistent store that have increasing/decreasing addresses. For example, a number of page writes could be done to locations 1, 5, 20, and 200 on the store. This avoids the time overhead associated with writing to locations that are unordered (e.g., locations 1, 200, 20, and 5).
  • the I/O journal filter 282 in one embodiment operates on a journal store that is implemented in a redundant fashion between the SSDs 54A, 54B, both of which also exhibit persistence. It should be appreciated that, while redundant and persistent stores are commonly utilized, other types of stores that do not exhibit redundancy or persistency can also be employed.
  • Each of the SSDs 54A, 54B has a copy of a journal 340, a data storage space of known length or capacity that stores the data associated with the IOBs and related metadata. Redundancy is provided by each of the SSDs 54A, 54B having a copy of the journal 340.
  • the operation of the I/O journal filter 282 is described with respect to a single copy of the journal 340, which may be referred as the journal 340, with the understanding that changes to one copy of the journal are also made to the other copy of the journal.
  • journal 340 has a data storage space of 640- Gigabytes. The storage space is divided into a plurality of 2-Megabyte journal page (JP) 342. Each journal page 342 has a journal page header 344 that identifies the journal page within the journal 340. The remainder of a journal page is available to be populated with a plurality of journal entries.
  • a journal entry (JE) 346 is comprised of a journal entry header (JEH) 348 that stores metadata related to the journal entry and a journal entry data field 350 capable of storing 4-kbytes of actual data associated with an IOB or other information that allows the actual data associated with the IOB to be reproduced.
  • the journal entry data field 350 is further divided into 8 512-byte journal block 351.
  • the journal entry header 348 is populated with the value for the layer LBA that is present in the LBA/PageNum field 226 of the IOB that provided the first 512-byte block in the journal entry data field and the values in the LayerlD, StorelD, and StoreLBA fields of the same IOB.
  • a one byte bit-mask is also present in the journal entry header 348 and is used to identify the 512-byte blocks that are in the journal entry data field 350. For example, if the LBA is 20 and the bit-mask is set to "10001000", LB As 20 and 24 are present in the journal entry data field 350.
  • journal table that maps the values in the LayerlD and LayerLBA fields of the IOB or journal entry header 348 to a particular journal page and journal entry.
  • FIG. 7 an example of a journal table 352 is illustrated.
  • the I/O journal filter 282 identifies IOBs in the execution queue 314 that have pending SCSI write-related commands (de-dup write and write), i.e., SCSI write-related commands that have not been fully addressed by an intervening filter.
  • the I/O journal filter 282 also identifies the currently active journal page and journal entry, i.e., the location in the journal 340 that is to be next in line to be populated with write-related data.
  • the currently active journal page could be journal page number "20" and the currently active journal entry could be journal entry "7".
  • the currently active journal entry either has no data in the journal entry data field or there is data in at least the first 512-byte journal block and one or more of the immediately following 512-byte journal blocks but not in all of the 512-byte journal blocks.
  • a "working" copy of the currently active journal page is located in the application memory of a storage processor. With respect to the "working" copy of the currently active journal page, the I/O journal filter 282 further determines if the first 512-byte block of the current journal entry has been written. If this is not the case, the I O journal filter 282 writes the next 512-byte block associated with an IOB into the first 512-byte block of the journal entry data field. If the IOB includes additional 512-byte blocks, these additional blocks (up to seven blocks) are also sequentially written into the current journal entry data field of the working copy.
  • the I/O journal filter 282 also writes the values from the LayerlD field 242, LBA/PageNum field 226, StorelD field 246, and StoreLBA field 248 into the journal entry header and sets the value in the bit-mask of the journal entry header to reflect the blocks that have been or will be loaded into the journal entry data field. For example, if the IOB includes five blocks of data, the I/O journal filter 282 would write the first of the five blocks of data into the first block of the journal data entry field and the other four blocks into the immediately following four blocks of the journal data entry field and establish the journal header data based on the first block of data moved into the journal data entry. In this example, the bit- mask would be set to "11 11 1000".
  • the I/O journal filter 282 uses the value of the layer ID in the journal entry header, the value of the LBA in the journal entry header, and the bit-mask in the journal entry header to determine the values for the LayerlD and the layer LBA that should go in the next available 512-byte block of the journal entry data field.
  • the I/O journal filter 282 would conclude that the block of data for layer id 0 and layer LBA 21 should go in the second block in the journal entry data field.
  • the calculated values for the layer id and layer LBA are compared to the actual layer id and layer LBA values associated with next block of data associated with the IOB. If there is a match, the next block of data associated with the IOB is written into the next available 512-byte block of the journal entry data field and the bit-mask is appropriately updated.
  • the I/O journal filter 282 establishes the 512-byte block of the IOB in the second 512-block of the journal entry data field. If there is not a match and the currently active journal entry is not the last journal entry for the currently active journal page, the currently active journal entry is incremented and the 512-byte block associated with the IOB is written in the first block of the new active journal entry.
  • the working copy of the active journal page is written to the actual journal 340 in the redundant and persistent store and a working copy of the next journal page is established in application memory.
  • the I O journal filter 282 will update the journal table 352. Specifically, the I/O journal filter 282 obtains the value from the LayerlD field 242 and the layer LBA value from the LBA/PageNum field 226. The I/O journal filter 282 determines if there is an entry in the journal table (e.g., journal table 352) that has the layer id and the layer LBA. If there is such an entry, the I/O journal filter 282 updates the journal page and journal entry fields with the currently active journal page and currently active journal entry. If there is not an entry, the I/O journal filter 282 creates and entry in the table and enters the layer ID, layer LBA, journal page, and journal entry values.
  • the journal table e.g., journal table 352
  • the I/O journal filter 282 operates with respect to IOBs in the execution queue 314 that have SCSI read-related commands (read) that have not been fully addressed by an intervening filter. More specifically, the I/O journal filter 282 obtains the value from the LayerlD field 242 and the layer LBA value from the LBA/PageNum field 226. The I O journal filter 282 determines if there is an entry in the journal table (e.g. journal table 352) that has the layer id and the layer LBA. If there is such an entry, the block(s) of data that are the subject of the read command are located in the journal at the journal page and journal entry specified for the entry in the journal table that has the noted layer id and layer LBA.
  • the journal table e.g. journal table 352
  • the I/O journal 282 proceeds to the specified journal entry, retrieves the LBA from the journal entry header, determines the difference between the requested layer LBA and the journal entry LBA to identify which of the 512-byte journal block(s) needs to be read.
  • the I/O journal 282 causes the relevant block(s) to then be read into memory store (e.g., memory store 52A or 52B) updates the DataSegmentVector field 240 to point to the location in memory store that contains the read blocks.
  • the I/O journal filter 282 places a "success" code in the error field 232 of the IOB and causes the IOB to start propagating up the filter stack 132, thereby indicating that the SCSI read command of the IOB has been completed.
  • the I/O journal filter 282 passes the IOB on down the filter stack 132.
  • the group of background filters 164 operates to place data on a data store with performance characteristics that are commensurate with the use of the data. For example, if a particular unit of data is frequently read and/or written, the group of background filters endeavor to place such data on a store with a high-performance characteristics (e.g., low latency, high throughput, and high IOPS). Conversely, if a particular unit of data is infrequently read and/or written, the group of background filters endeavor to place such data on a store with lower relative performance characteristics.
  • a high-performance characteristics e.g., low latency, high throughput, and high IOPS
  • the group of background filters 164 operates to move the unit of data in a manner that is speedy, conserves storage capacity, and has a relatively small impact on the processing of IOBs directly related to an initiator.
  • the group of background filters operate at the lowest criticality within the primary data storage system 28 or with an allowed latency that is significantly greater than the latency allowed in the foreground filters.
  • the background filters 164 operate in two contexts.
  • the first context involves the potential writing of data that is on one store to another store. In the background filters 164, such potential movements are accomplished using a super IOB that has a write-related SCSI block command and facilitates communications between the filters.
  • a super IOB is identical in form to IOB 182, except that the value of the PageMode field 224 is set to "on", which means that the values in the LBA/PageNum field 226 and the SectorCount/PageOffset field 228 now relate to pages and not blocks.
  • the second context involves the processing of an IOB that has a SCSI read-related command that has not yet been fully addressed by any of the filters in the filter stack 132 that have previously processed the IOB.
  • Operation of the background filters 164 with respect to operations that involve a super IOB is invoked by the I/O journal filter 282 indicating that a portion of the journal 340 is "dirty", i.e., has not been processed to determine whether data in the journal should be moved to a different store.
  • the actual percentage of the journal that is "dirty” is compared to a predetermined threshold value. If the actual percentage is less than the threshold percentage, operation of the background filters 164 is not invoked with respect to super IOBs. If the actual percentage of the journal that is "dirty” has a triggering relationship with respect to the threshold percentage (equals or exceeds, or only exceeds), operation of the background filters 164 is invoked for super IOBs.
  • the presence of the IOB in the execution queue 314 is detected and the operation of the background filters 164 is invoked.
  • the background filters 164 include a destage filter 370, advanced deduplication filter 372, page pool filter 374, store converter filter 376, and store statistics collection filter 378.
  • the destage filter 370 operates to move data between tiers of data stores with different characteristics and move the data so that the characteristics of the data reflect the characteristics of the store. In this regard, when the destage filter 370 is invoked because the percentage of the journal that is "dirty" has met some criteria, the destage filter 370 operates to determine if one or more pages of contiguous data blocks can be assembled from data blocks that typically are scattered throughout the journal. The destage filter 370 also makes a determination as to what should happen to any data blocks that cannot be assembled into a page.
  • the destage filter 370 If such a page can be assembled, the destage filter 370 generates a super IOB and passes the super IOB down the filter stack 132. The destage filter 370 further assesses whether each of the blocks that formed the page should, in addition to being the subject of the super IOB that will ultimately result in the blocks being written to another store, be persisted in the journal (i.e., whether a block is being read frequently enough to justify leaving the block in the journal). If two or more blocks are to be persisted in the journal, the destage filter 370 further assesses whether these blocks should remain in their current locations in the journal or be "compacted", i.e., consolidated into one or more consecutive journal entries. It should be appreciated that data for any specific layer and layer LBA may persist in multiple stores or tiers simultaneously.
  • the destage filter 370 operates to assess whether each such block has been resident in the journal for a period of time that exceeds a predefined threshold. If the threshold is exceeded, the destage filter 370 generates an IOB (not a super IOB) for the data block or group of contiguous blocks that is less than a page and passes the IOB down the filter stack 132. Further, the destage filter 370 assesses whether the block(s) should be persisted in the journal (i.e., whether the block(s) is being read frequently enough to justify leaving the block in the journal).
  • the destage filter 370 further assesses whether the blocks should remain in their current locations in the journal or be "compacted", i.e., consolidated into one or more consecutive journal entries. If the threshold is not exceeded, the destage filter 370 assesses whether the two or more blocks of data that are logically contiguous blocks that are separated from one another in journal but can be compacted into a single journal entry or journal page. If not, the blocks remain in their current locations in the journal.
  • the destage filter 370 determines if a page(s) can be assembled from the data blocks currently residing in the journal 340.
  • the destage filter 370 makes a working copy of the current journal table (e.g. journal table 352) and sorts the entries in the copy of the journal table by layer id and layer LBA.
  • the destage filter 370 analyzes the sorted journal table and, if necessary, the bit-masks in the headers of one or more journal entry headers 348 to determine if there is a layer with enough consecutive layer LB As of the data block size to equal a page.
  • the destage filter 370 assembles a working page in a memory store (memory store 52A or 52B). A super IOB is generated and the IOB is passed down the filter stack 132.
  • the destage filter 370 After the destage filter 370 assembles a page, the destage filter 370 builds a super IOB 182 and populates certain fields of the IOB 182 with values from or derived from the journal 340. Specifically, the destage filter 370 sets the command field 230 to block write command. If the data is a full page, then the destage filter 370 sets the PageMode field 224 of the IOB 182 as "on" to indicate that the IOB 182 initially relates to a page and not a block or blocks of data.
  • the "on" value in the PageMode field 224 also indicates that the values established or to be established in the LBA/PageNum field 226 and Sector Count/PageOffset field 228 are PageNum and PageOffset values and not LBA and SectorCount values.
  • the destage filter 370 uses data in the journal entry headers 348 to populate the LBA/PageNum field 226, Count/PageOffset field 228, LayerlD field 242, StorelD field 246, and StoreLBA field 248.
  • the destage filter 370 uses data in the journal entry headers 348 to establish values in the NumberOfDataSegments field 236 and the DataSegmentVector field 238.
  • the destage filter 370 places the data from the journal blocks 351 into the memory store (e.g., memory store 52A or 52B).
  • the destage filter 370 places the number of data segments that are established in the memory store into the NumberOfDataSegments field 236 and the address and length of each of the segments established in the memory into the DataSegmentVector field 238.
  • the destage filter 370 calculates a cyclic redundancy check (CRC) for each of the segments and places each of the CRC values in the DataCRCVector field 240. It should be appreciated that a data verification techniques other that CRC can be employed in place of CRC.
  • the value of the QoS Attributes field 244 is set to 0 or "lowest priority".
  • the destage filter 370 establishes such values in these fields.
  • the destage filter 370 also pushes an indication onto the IssuerStack field 252 of the IOB 182 that the destage filter 370 needs to do additional processing when the IOB is propagating up the filter stack 132.
  • the destage filter 370 also updates a cache entry (CE) in a cache table for each journal entry that contributed one or more blocks to the page to indicate that the data associated with the journal entry is being destaged, i.e., is now the subject of a super IOB that will result in the data being written to a different data store. More specifically, a state bit mask in the CE is updated to indicate that the data associated with the journal entry is being destaged.
  • CE cache entry
  • the destage filter 370 makes a determination of whether or not to persist the data block on the journal 340.
  • the destage filter 370 obtains statistical data from the statistics database 168 for the layer ID and layer LBA associated with the block. If the statistical data indicates that the data block is not being frequently read, the destage filter 370 removes the entry for the layer ID and layer LBA in the journal table (e.g., journal table 352) and updates the state bit mask in the related CE to indicate that the data block has been evicted from the journal 340. This effectively frees up the JE for the data block for use by the I/O journal filter 282.
  • the destage filter 370 makes a determination as to whether to leave the data block in its current location or compact the data block with other data blocks that are being persisted. To make this determination, the destage filter 370 assesses whether the journal page that contains the data block is sparsely populated or not. If the journal page is sparsely populated and there is at least one other data block associated with another sparsely populated journal page, the destage filter 370 compacts the two data blocks into one journal page, thereby freeing up one journal page for use by the I/O journal filter 282. If the journal page is not sparsely populated, the data block is allowed to remain in its current location in the journal 340.
  • the destage filter 370 determines that: (a) a page could not be assembled from the data blocks resident in the journal 340 when the destage filter 370 began processing the journal 340 ("unpageable data blocks") or (b) the journal had data blocks that could be assembled into a page (“pageable data blocks”) and unpageable data blocks, the destage filter 370 processes each of the unpageable data blocks in the journal to assess how long the data block has been resident in the journal 340.
  • the destage filter 370 obtains the current time, obtains the "write” time from a time stamp field in the CE for the layer ID and the layer LBA that relates to the data block to determine when the data block was written into the journal 340, and determines the difference between the current time and the "write" time.
  • the destage filter 370 creates an IOB (not a super IOB) for the data block and any contiguous data blocks in a similar fashion to that noted for the super IOB but with a PageMode value set to "off and passes the IOB on down the filter stack 132. Additionally, the destage filter 370 makes a determination of whether or not to persist the data block on the journal 340. In this regard, the destage filter 370 obtains statistical data from the statistics database 168 for the layer ID and layer LBA associated with the block.
  • the destage filter 370 removes the entry for the layer ID and layer LBA in the journal table (e.g., journal table 352) and updates the state bit mask in the related CE to indicate that the data block has been evicted from the journal 340. This effectively frees up the JE for the data block for use by the I/O journal filter 282. If the statistical data indicates that the data block is being frequently read, the destage filter 370 makes a determination as to whether to leave the data block in its current location or compact the data block with other data blocks that are being persisted. To make this determination, the destage filter 370 assesses whether the journal page that contains the data block is sparsely populated or not.
  • the destage filter 370 compacts the two data blocks into one journal page, thereby freeing up one journal page for use by the I O journal filter 282. If the journal page is not sparsely populated, the data block is allowed to remain in its current location in the journal 340.
  • the destage filter 370 makes a determination as to whether to leave the data block in its current location or compact the data block with other data blocks that are being persisted. To make this determination, the destage filter 370 assesses whether the journal page that contains the data block is sparsely populated or not. If the journal page is sparsely populated and there is at least one other data block associated with another sparsely populated journal page, the destage filter 370 compacts the two data blocks into one journal page, thereby freeing up one j ournal page for use by the I O j ournal filter 282. If the j ournal page is not sparsely populated, the data block is allowed to remain in its current location in the journal 340.
  • the destage filter 370 queries the statistics database 168 to determine if the system has sufficient resources to process the destage. If the system does have sufficient resources, the destage filter 370 places an "In" time in the In Time Stamp field 250 that reflects the point in time when or about when the destage filter 370 passes the IOB 182 on down the filter stack 132. If the system does not have resources to process the destage IOB, then the destage filter pauses and then tries the stats database query again.
  • the advanced de-duplication filter 372 operates to preserve storage capacity at the primary data storage system 28 by preventing blocks of data associated with a super IOB that are commonly written to the primary data storage system 28 and do not have a readily calculable pattern from being written multiple times such that each writing of the page consumes additional storage capacity.
  • the advanced de-duplication filter 372 maintains a super dictionary table that is capable of holding a number of entries that is greater than the number of entries that the dictionary table associated with the dictionary deduplication filter 280 utilizes.
  • Each non-null entry in the super dictionary table includes, for a page associated with a super IOB, a value for each of a cyclic redundancy check (CRC) for the page, a layer ID, PageNum, a StorelD, and StoreLBA.
  • CRC cyclic redundancy check
  • the CRC is a number that is calculated using the data in a page and representative of the data in a page but not necessarily a unique number relative to the data in the page, i.e., there is the possibility that two pages with different data have the same CRC. Nonetheless, if two pages of data do have the same CRC, there is a distinct possibility that the two pages do have the same data. It should be appreciated that hashes, checksums, and the like can be used in lieu of a CRC to identify pages that have potentially identical data.
  • the advanced deduplication filter 372 calculates a CRC for the page located in a memory store (memory store 52A or 52B) due to the operation of the destage filter 370.
  • the advanced deduplication filter 372 enters the calculated CRC value and the values from the LayerlD field 242, PageNum field 226, StorelD field 246, and StoreLBA field 248 in the super dictionary table.
  • the advanced deduplication filter 372 determines if there is another entry in the super dictionary table that has the same CRC value, the same value for the StorelD, and the value for the StorelD corresponds to a memory store.
  • Two entries in the super dictionary table with the same CRC value are potentially identical pages.
  • Two entries in the super dictionary table that also each has a value for the StorelD that corresponds to a memory store (which is a high speed memory) can be compared to one another very quickly.
  • the data associated with the write IOB and the dictionary entry are both in memory store 52A or 52B, the fastest type of store in the illustrated system. If there is another entry in the super dictionary table that has the same CRC value and a value for the StorelD that corresponds to a memory store, the advanced deduplication filter compares the two pages to one another.
  • the advanced deduplication filter 372 changes the value in the command field 230 of the super IOB from a write to a de-dup write, adjusts the values in the StorelD field 246 and StoreLBA field 248, and passes the super IOB on down the filter stack 132.
  • the advanced deduplication filter 372 increments a page counter that is used to determine whether the identical page is being written commonly or frequently enough to warrant identifying the page as being appropriate for use in the dictionary table used by the dictionary deduplication filter 280 in the group of foreground filters 162. If the page satisfies the test for inclusion in the dictionary table, the advanced deduplication filter obtains the portion of the page (e.g., the second 64-bytes in the page) that is associated with each of the non-null entries in the dictionary table. If the portion of the page is unique relative to each of the portions of the pages associated with the other entries, the page is added to the dictionary table.
  • a page counter that is used to determine whether the identical page is being written commonly or frequently enough to warrant identifying the page as being appropriate for use in the dictionary table used by the dictionary deduplication filter 280 in the group of foreground filters 162. If the page satisfies the test for inclusion in the dictionary table, the advanced deduplication filter obtains the portion of the page (e.g.,
  • the advanced deduplication filter 372 operates to identify a portion of each of the pages in the dictionary table that is unique and updates the entire dictionary table accordingly. If a portion of each of the pages in the dictionary table that is unique cannot be identified, the page is not added to the dictionary table.
  • the advanced deduplication filter 372 proceeds to assess the impact of considering whether other entries in the super dictionary table having the same CRC are duplicates of the page associated with the super IOB. Specifically, the advanced deduplication filter 372 queries the statistics database 168 to determine if the QoS goals are currently being achieved or nearly achieved (a "headroom" calculation). If the impact is acceptable, the advanced deduplication filter 372 causes the page that is at the location identified by the values in the StorelD and StoreLBA fields in the super dictionary table to be read into a memory store for comparison to the page associated with the super IOB currently in the memory store.
  • the comparison proceeds in substantially the same fashion as described above when the two pages were both in memory store when the processing of the super IOB by the advanced deduplication filter 372 began. If the impact is not acceptable, the advanced deduplication filter 372 passes the super IOB on down the filter stack 132. If there is no entry in the super dictionary table that has the same CRC, the advanced deduplication filter 372 passes the super IOB on down the filter stack 132.
  • the advanced deduplication filter 372 deletes the entry in the super dictionary table that has the layer ID and the PageNum values set forth in the IOB.
  • the entry is deleted because the write command associated with the IOB will be subsequently executed and likely change the CRC for the page with which the data block(s) that are the subject of the write command are associated. As such, the current CRC for the page will no longer be valid and useable for assessing whether there is a page that is the subject of a super IOB should be deduplicated.
  • the advanced deduplication filter 372 passes the IOB on down the filter stack 132.
  • the advanced deduplication filter 372 operates on an IOB having a SCSI read-related command that need not relate to a page to determine if the data associated with the identified layer id and LBA is data that has been previously de-duplicated in the processing of an IOB with a SCSI write-related command relating to the same layer id and LBA. More specifically, the advanced deduplication filter 372 obtains the value in the StorelD field 246 and determines if the value is currently associated with the dictionary store 322. If the value is currently associated with the dictionary store 322, the advanced deduplication filter 372 then places the data from the dictionary store 322 into the memory store (e.g., memory store 52A or 52B).
  • the memory store e.g., memory store 52A or 52B
  • the advanced deduplication filter 372 places the number of data segments that are established in the memory store into the NumberOfDataSegments field 236 and the address and length of each of the segments established in the memory into the DataSegmentVector field 238. Further, the advanced deduplication filter 372 updates the value in the DataSegmentVector field to point to the address in the memory store (e.g., memory store 52A or 52B) that has the copy of the dictionary page and, more specifically, to point the first block of the page that has the first block to which the SCSI read command relates.
  • the memory store e.g., memory store 52A or 52B
  • the advanced deduplication filter 372 places a "success" code in the error field 232 and causes the IOB to start propagating up the filter stack 132, thereby indicating that the SCSI read-related command of the IOB has been completed. If the value in the StorelD field 246 is not currently associated with the dictionary store 322, the IOB is passed down the filter stack 132 for further processing.
  • Iob.StorelD OneCandidate->LocationStore
  • Iob.StoreLBA OneCandidate->LocationLBA
  • Iob.StorelD OneCandidate->LocationStore
  • Iob.StoreLBA OneCandidate->LocationLBA
  • Iob.StorelD OneCandidate->LocationStore
  • Iob.StoreLBA OneCandidate->LocationLBA
  • TestBuffer MemroyGetDataBuffer (TestCandidate->LocationLBA) CmpBuffer ( lob, TestBuffer )
  • TestBuffer SSDGetDataBuffer (TestCandidate->LocationLBA) CmpBuffer ( lob, TestBuffer )
  • TestBuffer SAS (TestCandidate->LocationLBA)
  • the page pool filter 374 operates to allocate storage space on the stores associated with the primary data storage system 28 other than a store that is non-persistent and any portion of a store that is not dedicated to a journal as needed. More specifically, the page pool filter 374 maintains a store map for each store for which the filter can allocate storage that identifies all of the storage pages on the store and indicates whether or not each such storage page has been allocated. Additionally, the page pool filter 374 maintains a layer-store table 410 with each entry in the table mapping a layer ID and layer LBA to a StorelD and StoreLBA. The table also indicates whether the data at a particular StorelD and StoreLBA is shared by more than one layer ID, layer LBA.
  • This indication is referred to as a ref-count, with a ref-count of 1 indicating that the data at the location specified by the StorelD and StoreLBA is only associated with one layer ID, layer LBA.
  • a ref-count that is greater than 1 indicates that the data at the location specified by the StorelD and Store LBA is associated with more than one layer ID, layerLBA.
  • the page pool filter 374 operates on a received IOB to determine if the received IOB is an IOB or a super IOB. More specifically, the page pool filter 374 obtains the value in the PageMode field 224 of the received IOB. If the value is "yes", the received IOB is a super IOB, i.e., embodies a write-related command that involves a page of data.
  • the page pool filter 374 determines whether the command in the command field 230 is a write command or a dedup write command. If the command is a write command, the page pool filter 374 obtains the values in the LayerlD field 242 and the LBA/PageNum field 226 and determines whether there is an entry in the layer- store table 410. If there is no entry in the layer-store table 410 with the specified layer ID and layer LBA values, the page of data for the specified layer ID and layer LBA has not been previously written to any of the stores for which the page pool filter 374 allocates space.
  • the page pool filter 374 interrogates the store map(s) to identify a page of space on the related store to which the page of data can be efficiently written. With respect to an identified page, the page pool filter 374 determines the values for the StorelD and StoreLBA. The page pool filter 374 allocates the page to the layer ID and layer LBA. In this regard, the page pool filter 374 updates the layer-store table to include an entry with the values for the layer ID, layer LBA, StorelD and StoreLBA and stores the updated store map. Further, the page pool filter 374 sets the ref-count field in the entry to 1 to indicate that the data to be established beginning at the location specified by the StorelD and StoreLBA values is currently associated with only one layer ID and layer LBA. The page pool filter 374 updates the StorelD field 246 and StoreLBA field 248 in the IOB with the StorelD and StoreLBA values of the allocated storage. The updated super IOB is then passed down the filter stack 132.
  • the page pool filter 374 determines if the data is shared, i.e., associated with another layer ID and layer LBA values. In this regard, the page pool filter 374 determines if the ref-count field in the entry in the layer-store table 410 for the layer ID and layer LBA in the super IOB is 1. If the ref-count is 1, the data at the location specified by the StorelD and StoreLBA values in the table is not shared. In this case, the values for the StorelD and StoreLBA in the table are respectively loaded into the StorelD field 246 and StoreLBA field 248. The updated super IOB is then passed on down the filter stack 132.
  • the page pool filter 374 If the ref-count is greater than 1, the data at the location specified by the StorelD and StoreLBA for the entry in the layer-store table 410 is shared with at least one other layer ID and layer LBA. In this case, because the data at the location is shared and the IOB involves the writing of data that is different than the data currently at the location, the page pool filter 374 must allocate new space on a store for the page of data associated with the super IOB. In this regard, the page pool filter 374 proceeds substantially as noted with respect to the situation in which there was no entry in the layer-store table 410 with the specified layer ID and layer LBA values. Further, the page pool filter 374 also decrements the ref-counts.
  • the page pool filter 374 establishes a new entry in the layer-store table 410 and populates the entry with the values from the LayerlD field 242, LBA/PageNum 226 field 226, StorelD field 246, and the StoreLBA field 248 from the super IOB. In this instance, the values in the StorelD field 246 and the StoreLBA field 248 were previously established by the advanced deduplication filter 372. Further, the page pool filter 374 identifies the other entries in the layer-store table 410 that have the same value for the StorelD and StoreLBA.
  • the page pool filter 374 With respect to each of these entries in the layer-store table 410 the ref-count value is incremented. The page pool filter 374 also establishes this incremented ref-count value in the new entry in the layer- store filter. The processing with respect to this super IOB is now complete. Consequently, the page pool filter 374 places a "success" code in the error code field 232 and causes the IOB to start propagating up the filter stack 132.
  • the page pool filter 374 determines whether the command in the command field 230 is a write command or a read command. If the command is a write command, the page pool filter 374 obtains the values in the LayerlD field 242 and the LBA/PageNum field 226 and determines whether there is an entry in the layer-store table 410. If there is no entry in the layer-store table 410 with the specified layer ID and layer LBA values, the block(s) of data for the specified layer ID and layer LBA has not been previously written to any of the stores for which the page pool filter 374 allocates space.
  • the page pool filter 374 interrogates the store map(s) to identify a page of space on the related store to which the block(s) of data can be efficiently written. With respect to an identified page, the page pool filter 374 determines the values for the StorelD and StoreLBA. The page pool filter 374 allocates the page to the layer ID and layer LBA. In this regard, the page pool filter 374 updates the layer-store table 410 to include an entry with the values for the layer ID, layer LBA, StorelD and StoreLBA and stores the updated store map.
  • the page pool filter 374 sets the ref-count field in the entry to 1 to indicate that the data to be established beginning at the location specified by the StorelD and StoreLBA values is currently associated with only one layer ID and layer LBA.
  • the page pool filter 374 updates the StorelD field 246 and StoreLBA field 248 in the IOB with the StorelD and StoreLBA values of the allocated storage. The update IOB is then passed down the filter stack 132.
  • the page pool filter 374 determines if the data is shared, i.e., associated with another layer ID and layer LBA. In this regard, the page pool filter 374 determines if the ref-count field in the entry in the layer-store table 410 for the layer ID and layer LBA in the IOB is 1. If the ref-count is 1 , the data at the location specified by the StorelD and StoreLBA values in the layer-store table 410 is not shared.
  • the values for the StorelD and StoreLBA in the layer-store table 410 are respectively loaded into the StorelD field 246 and StoreLBA field 248.
  • the super IOB is then passed on down the filter stack 132.
  • the ref-count is greater than 1, the data at the location specified by the StorelD and StoreLBA for the entry in the layer-store table 410 is shared with at least one other layer ID and layer LBA. In this case, because the data at the location is shared and the IOB involves the writing of data that is different than the data currently at the location, the page pool filter 374 must allocate new space on a store for the page of data associated with the super IOB.
  • the page pool filter 374 must build the page that is to be written to the newly allocated space. Consequently, the page pool filter 374 reads the page that is at the location specified by the current StorelD and StoreLBA in the layer-store table 410 into a memory store (e.g., memory stores 52A or 52B) and modifies the page to include the block(s) that are associated with the IOB.
  • a memory store e.g., memory stores 52A or 52B
  • the page pool filter 374 establishes a new entry in the layer-store table 410 and enters the values from the LayerlD field 242 and LBA/PageNum field 226 of the IOB into the new entry in the table. Further, the StorelD and StoreLBA values for the newly allocated space are also placed in the new entry. The ref-count for the new entry is set to 1 to indicate that the page is not shared with any other layer ID and layer LBA.
  • the page pool filter 374 updates the values of the StorelD field 246 and the StoreLBA field 248 in the IOB to reflect the StorelD and StoreLBA for the newly allocated space. Further, the page pool filter 374 updates the DataSegmentVector 240 in the IOB to indicate the location of the modified page in the memory store. The updated IOB is then passed down the filter stack 132.
  • the page pool filter 374 uses the values from the LayerlD field 242 and the LBA/PageNum field 226 to identify the entry in the layer- store table 410 that relates to the data that is to be read.
  • the value in the LBA/PageNum field 226 relates to an LBA and not a page.
  • the page pool filter 374 accomplishes the conversion by masking off certain bits of the LBA value.
  • the layer ID and PageNum values are then used to identify the entry in the layer-store table 410 relating to the data that is the subject of the read command.
  • the page pool filter 374 retrieves the values for the StorelD and StoreLBA associated with the entry in the layer-store table 410 and loads these values into the StorelD field 246 and StoreLBA fields 248 of the IOB.
  • the updated IOB is then passed down the filter stack 132.
  • the store converter filter 376 processes super IOBs and IOBs so as to generate an element specific IOB(s), i.e., the command(s) that are needed to actually perform the read or write of the data associated with the super IOB or IOB.
  • an element specific IOB(s) i.e., the command(s) that are needed to actually perform the read or write of the data associated with the super IOB or IOB.
  • a particular store has data transfer requirements, a data redundancy attribute, and a path redundancy attribute.
  • the store converter filter 376 processes super IOBs and IOBs to produce the element specific IOB(s) with the command(s) to the store that satisfy the data transfer requirements of the store, preserve the data redundancy attribute of the store, and preserve the path redundancy attribute of the store.
  • the store converter filter 376 interrogates a store table to obtain the size of a write-related data transfer that the store accommodates. If the size of the data transfer accommodated by the store is equal to a page, the store converter filter 376 generates the element specific IOB with the command(s) necessary to write the page of data associated with the super IOB to the store.
  • the store converter filter 376 With respect to a super IOB with a write-related command, if the size of the data transfer accommodated by the store is greater than a page, the store converter filter 376 generates the element specific IOB(s) with the command(s) necessary to: (a) read the current greater portion of data that is on the store and that includes the location at which the page is to be written, (b) modify the read current greater portion of data to include the page of data associated with the super IOB, and (c) write the modified greater portion of data to the store.
  • the store converter filter 376 generates the commands necessary to: (a) read the current 4- megabyte chunk of data on the store that includes the location at which the page associated with the super IOB is to be written, (b) modify the read 4-megabyte chunk to include the page associated with the super IOB, and (c) write the modified 4-megabyte chunk to the store.
  • the store converter filter 376 divides the page of data associated with the super IOB into whatever size chunks of data are required by the store and generates the element specific IOB(s) with the command(s) for transferring these chunks of data to the store. For instance, if a store requires that data to be written in 512-byte chunks, the store converter filter 378 divides the 2-megabyte page associated with the super IOB into 4096 512-byte chunks and generates the command(s) for writing each of the 4096 512-byte chunks to the store.
  • the store converter filter 376 (a) divides the page into one or more chunks of the size required by the store and generates the command(s) for writing each of these chunks to the store and (b) with respect to the remaining data that is less than the size of data transfer accommodated by the store, produces the read, modify, write commands previously described for writing the data to the store.
  • the store converter filter 376 operates in substantially the same fashion as noted with respect to a super IOB, except that the size of the block or blocks of data that are the subject of the IOB rather than a page are compared to the size of the data transfer accommodated by the store.
  • the store converter filter 376 also interrogates the store table to determine the value of the data redundancy attribute associated with the store, performs any calculations that are associated with satisfying this attribute for the store, and generates or modifies the element specific IOB so as to implement the data redundancy. For example, if a store is comprised of a RAID-6 element, the store converter filter 376 engages in the parity calculations that are needed for use with a store that includes such an element and modifies the element specific IOB accordingly. As another example, if the store includes two elements that are mirrored to provide data redundancy, the store converter filter 376 modifies the element specific IOB to include the command(s) needed for implementing the mirroring.
  • the store converter filter 376 further interrogates the store table to determine the value of the path redundancy attribute associated with the store.
  • the store converter filter 376 interrogates a configuration table for the primary data storage system 28 that provides the physical layout of the level and the characteristics of the various elements at the level. For example, the configuration table identifies each store, the number of I/O ports associated with each store, the status of the ports, identifies the switches in the store and the status of the switches etc.
  • the store converter filter 376 generates or modifies the element specific IOB to provide the necessary information for routing the data from its current location in the primary data storage system 28 (e.g., the memory store) to the store.
  • the store converter filter 376 interrogates a store table to obtain the size of a read-related data transfer that the store accommodates. If the size of the read data transfer accommodated by the store is equal to the size of the data that is the subject of the IOB, the store converter filter 376 generates the element specific IOB with the command(s) necessary to read the data associated with the IOB from the store.
  • the store converter filter 376 If the size of a data transfer accommodated by the store is greater than size of the data that is the subject of the IOB, the store converter filter 376 generates the element specific IOB with the command(s) necessary to read the current greater portion of data that is on the store and that includes the location with the data that is the subject of the IOB into the memory store. The store converter filter 376 then updates the value in the DataSegmentVector field to point to the address in the memory store (e.g., memory store 52A or 52B) that has the copy of the page and, more specifically, to point the first block of the page that has the first block to which the SCSI read command relates.
  • the memory store e.g., memory store 52A or 52B
  • the store converter filter 376 determines the number of data transfers that will be necessary to transfer data of the size specified in the IOB and generates the element specific IOB(s) with the command(s) for conducting the calculated number of reads from the store.
  • the store converter filter 376 determines the number of data transfers that will be necessary to transfer data of the size specified in the IOB and generates the element specific IOB(s) with the command(s) for conducting the calculated number of reads from the store and (b) with respect to the remaining data that is less than the size of data transfer accommodated by the store, generates or modifies the element specific IOB to include the command(s) necessary to read the portion of data that is on the store that is of a greater size than the remaining data but includes the location with the remaining data.
  • the store converter filter 376 accesses a hardware state table to determine which path(s) and element(s) to which the element specific IOB should be sent.
  • the store stats collection filter 378 operates to collect certain store and element related data/statistical information for each IOB passed to the store stats collection filter 378 from the store converter filter 376 when the IOB is going down the filter stack 132.
  • the store stats collection filter 378 processes the IOB 182 to obtain the store id from the Storeld field 246, the element id from the ElementID field 256, the sector count from the SectorCount/PageOffset field 228, and the "In" time stamp value from the In Time Stamp field 250.
  • the store stats collection filter 378 also obtains the current time from the operating system.
  • the store stats collection filter 378 uses the value of the "In” Time Stamp and the current time to calculate the latency that the IOB has experienced between when the "In” Time Stamp value was established in the destage filter 370 and when the current time is obtained by the store stats collection filter 378 (hereinafter referred as "first latency").
  • the store stats collection filter 378 communicates with the statistics database 168 so as to: (a) update a table for the store that is maintained in the database to reflect that an IOB associated with the store will be processed that has the sector size obtained from the IOB and that the IOB has experienced the calculated first latency and (b) update a table for the element that is maintained in the database to reflect that an IOB associated with the element will be processed that has the sector size obtained from the IOB and that the IOB has experienced the calculated first latency.
  • the store stats collection filter 378 also pushes an indication onto the IssuerStack field 252 of the IOB 182 that the store stats collection filter 378 needs to do additional processing when the IOB is propagating up the filter stack 132. Further, the store stats collection filter 378 also pushes the current time onto the XtraContextStack field 254.
  • the store stats collection filter 378 obtains the time from the XtraContextStack field 254 (which is no longer the current time), obtains the "new" current time, and calculates a second latency, i.e., the elapsed time between when the time value was obtained that was pushed onto the XtraContextStack field 254 and the IOB was propagating down the filter stack 132 and the when the "new" current time was obtained.
  • the store stats collection filter 378 updates the store and element tables in the statistics database 168 with the second latency value.
  • the storage hardware driver 380 controls a SCSI card so as to produce the electrical signals needed to receive a message, such as SCSI block result, and transmit a message, such as a SCSI block request.
  • the storage hardware driver 380 assures the addressing of packets associated with a message. With respect to received packets, the storage hardware driver 380 confirms that each of the received messages does, in fact, belong to the SCSI card. With respect to messages that are to be transmitted, the storage hardware driver 380 assures that the each message is appropriately addressed so that the message gets to the desired element. With respect to a received message, the storage hardware driver 380 also recognizes the packet as requiring further routing back up the filter stack 132. The storage hardware driver 380 also performs other processing in accordance with the protocols, e.g., ordering packets, checksum etc.
  • the storage hardware driver 380 operates to process block commands, i.e., commands that relate to the reading of a block data from or writing of a block data to a storage medium.
  • the storage hardware driver 380 can be adapted to operate with storage hardware other that SCSI cards.
  • a number of functions noted with respect to the primary data storage system 28 can be realized with a primary data storage system having a single storage processor and a single data store and primary data storage systems having more elements than noted with respect to the primary data storage system 28.
  • the tiering function described with respect to I/O journal filter and the destage filter can be practiced in a primary data system with two data stores having different performance characteristics.
  • the QoS function described with respect to the QoS filter can be practiced in a primary data storage system that has a single data store where there are two are more volumes associated with the store.
  • the de-duplication function can be practiced in a primary data storage system with a single data store.
  • a primary data storage system can employ a filter stack with a fewer number or greater number of filters than are in the filter stack 132. For instance, in a primary data storage system that is only going to service a single volume, a filter stack can be employed that omits a QoS filter. Additionally, a filter stack can be employed in which the order of filters in the stack are different than in filter stack 132. For instance, a filter stack could be employed in which the an I/O journal filter preceded a the dictionary deduplication filter.
  • a tier is a group of stores that have similar characteristics such as throughput, latency, capacity, path redundancy, data redundancy, and atomic block size (i.e., the smallest individually addressable block of a store) or a store with a defined set of such characteristics.
  • memory store 52A and 52B comprise a tier
  • RAID disk array 56A and 56B comprise a different tier
  • SSDs 54A and 54B comprise yet another tier.
  • One tier can differ from another tier in one characteristic or multiple characteristics. For instance, a particular tier may have specific latency and throughput characteristics while another tier may have the same latency but a different throughput characteristic.
  • a tiering storage system is a storage system that attempts to match the access pattern relating to a block of data in the system to the tier having the most appropriate or compatible characteristics.
  • the QoS filter 274 evaluates an IOB and volume, criticality, and hardware statistics from the statistics database 168 to determine the most compatible and available tier(s) for the blocks of data relating to an IOB.
  • the QoS filter 274 updates the AllowedStores field 260B of the IOB with the identified tier(s). It should be appreciated that the AllowedStores field 260B can be implemented as a bitmask and the QoS filter 274 can indicate in the bitmask that an IOB should skip a tier.
  • the QoS filter 274 might indicate that the write data associated with the IOB be written to the RAID disk array 56A or 56B instead of the SSDs 54A or 54B, which are in a higher tier than the RAID disk arrays 56A, 56B.
  • the pattern de-duplication filter 278 and the calculation engine 320 implement a tier- 1 (the fastest tier, but with a limited capacity) functionality in the illustrated primary data storage system 28.
  • the pattern de-duplication filter 278 operates to identify and respond to IOBs that contain blocks of data capable of being stored or retrieved from the calculation engine 320 or other similar engines.
  • the calculation engine 320 provides a CPU store for storing and retrieving blocks of data that are readily calculable.
  • the calculation engine 320 is implemented by using a CPU and a limited amount of high speed memory to store and retrieve blocks of data.
  • the calculation engine has a block size characteristic of 512 bytes (the smallest of any tier).
  • the calculation engine 320 has the lowest latency and highest bandwidth of the stores illustrated. It should be appreciated that the calculation engine 320 could be realized using specialized hardware such as a DMA engine or an MMX processor.
  • the dictionary de-duplication filter 280 and the dictionary store 322 implement a tier-2 (slower than tier- 1 but with greater capacity than tier- 1 ) functionality.
  • the dictionary de-duplication filter 280 operates to identify and respond to IOBs that contain blocks of data that are identical to the blocks of data stored in the dictionary store 322.
  • the dictionary store 322 provides a dictionary table and a memory store 52A or 52B for storing and retrieving blocks of data which are not readily calculable.
  • the dictionary store 322 has a block size characteristic of 2MB.
  • the I/O journal filter 282 and the SSDs 54A and 54B implement a tier-3 (slower than tier-2 but with greater capacity than tier-2) functionality.
  • the I/O journal filter 282 operates to identify and respond to IOBs that the filters above in the filter stack 132 have not fully processed.
  • the I/O journal filter 282 stores blocks of data to and retrieve blocks of data from the SSDs 54A and 54B based upon the characteristics of the SSDs 54A and 54B (e.g. atomic block size, performance, throughput, IOPs, persistence, and redundancy).
  • the SSDs 54A and 54B each provide a persistent store for storing blocks of data.
  • the SSDs 54A and 54B each have an atomic block size characteristic of 4KB.
  • the destage filter 370 is responsible for movement of blocks of data between two tiers.
  • the destage filter 370 decides when blocks of data relating to an IOB should be copied, moved, or cleared relative to multiple tiers (in the illustrated system 28, the tier-3 SSDs 54A or 54B and the tier-4 RAID disk array 56A or 56B).
  • the destage filter 370 uses the characteristics of the source and destination tiers to accommodate the different tier requirements. For example, the SSDs 54A and 54B require atomic block accesses to be 4KB in size while the RAID disk array 56A and 56B require atomic block accesses to be 2MB (page size).
  • destage filter 370 executes a multitude of reads from the SSDs 54A or 54B in 4KB chunks that coalesce in high speed memory until 2MB have been read. The destage filter 370 then executes a write command to the RAID disk array 56A or 56B with the 2MB that is now in high speed memory. Likewise, the destage filter 370 evaluates other characteristics of the various stores and accommodates the characteristic strengths and attempts to avoid the characteristic weaknesses. For example, the RAID disk array 56A or 56B has a seek penalty. Due to this penalty, the destage filter 370 processes IOBs in a fashion to limit or reduce this seek penalty impact. The ability of destage filter 370 to accommodate various characteristics of different stores enables more efficient use of resources.
  • the atomic block size of the SSDs 54A and 54B is smaller than the atomic block size of the RAID disk array 56A or 56B which allows the SSDs 54A and 54B to contain smaller segments of more frequently accessed blocks of data and not require the SSDs 54A and 54B to hold blocks of data that are adjacent to the frequently accessed blocks of data. In effect this is more efficient use of the SSDs 54A and 54B.
  • the destage filter 370 can also copy blocks of data between tiers so as to maintain a block of data in multiple tiers and thus increasing redundancy associated with the block of data. This also allows the block of data that is located in multiple tiers to be "fast reused". Fast reuse occurs when a tier includes a copy of a block(s) (i.e., there is another copy in another tier) and it is necessary to make space in the tier for a block or blocks of data associated with a different IOB command. In this case, the copy of the block(s) in the tier can be deleted/written over to make space for the block(s) associated with the different IOB command.
  • the destage filter 370 endeavors to match a block or blocks of related data to the tier that is appropriate for the access pattern associated with the block or blocks of related data. To accomplish this, the destage filter 370 accesses the statistics database 168 to acquire historical statistics related to the volume with which the data block or related data blocks are associated and evaluates those statistics to detect trends in the access pattern. For example, if the initiator access pattern is a streaming video (a trend represented by a sequence of consecutive IOBs), the destage filter 370 would likely direct the blocks of data to the tier containing the RAID disk array 56A or 56B because the RAID disk array 56A or 56B is more efficient than other tiers in processing large, contiguous blocks of data.
  • the destage filter 370 endeavors to maintain the blocks of data in a tier such as SSDs 54A and 54B because this tier has a smaller seek latency penalty relative to the other tiers in the system.
  • the advanced de-duplication filter 372 provides movement of blocks of data between tier-4 and tier-2. More specifically, advanced de-duplication filter 372 uses the super dictionary table to determine when a group of contiguous blocks of data that constitute a page is frequently accessed. If a page is accessed more frequently than other pages active in the dictionary table, then the advanced de-duplication filter 372 identifies that page as a candidate for movement to tier-2. The advanced de-duplication filter 372 subsequently coordinates with the dictionary de-duplication filter 280 to update the dictionary table with the candidate page.
  • the page pool filter 374 and the RAID disk array 56A or 56B implement a tier 4 (slower than tier-3 but with greater capacity than tier-3) functionality.
  • the page pool filter 374 operates to store and retrieve blocks of data from RAID disk array 56A and 56B considering the characteristics of RAID disk array 56A and 56B.
  • tiering functionality can be implement with other combinations of filters and stores. It should also be appreciated that other filter stack 132 layouts could generate different tier assignments than those listed above. Additional storage types such as the cloud storage provider 64 or tape stores would likely involve the filter stack 132 adding additional filters or re-arranging the order of the filters in such a way as to accommodate the characteristics of any new tier employing one or more of these types of stores. Further, as faster stores become available, these faster stores can be used to implement a tier that is faster than the memory that constitutes the tier- 1 in the illustrated system.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Human Computer Interaction (AREA)
  • Computer Hardware Design (AREA)
  • Quality & Reliability (AREA)
  • Computer Security & Cryptography (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

L'invention porte sur un système de stockage de données primaires pour une utilisation dans un réseau informatique dans lequel un réseau permet à des ordinateurs d'utilisateurs de transférer des données vers/depuis le système de stockage de données primaires. Dans un mode de réalisation, le système de stockage de données primaires met en place une stratégie de hiérarchisation pour déplacer des données entre des stockages dotés de caractéristiques différentes de façon à faire correspondre les caractéristiques des données aux caractéristiques d'un des stockages.
PCT/US2012/063995 2011-11-07 2012-11-07 Système de stockage de données primaires doté d'une hiérarchisation de données WO2013070803A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US201161556815P 2011-11-07 2011-11-07
US61/556,815 2011-11-07

Publications (1)

Publication Number Publication Date
WO2013070803A1 true WO2013070803A1 (fr) 2013-05-16

Family

ID=47258103

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2012/063995 WO2013070803A1 (fr) 2011-11-07 2012-11-07 Système de stockage de données primaires doté d'une hiérarchisation de données

Country Status (2)

Country Link
US (2) US10216651B2 (fr)
WO (1) WO2013070803A1 (fr)

Families Citing this family (34)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9141288B2 (en) * 2012-07-26 2015-09-22 Hewlett-Packard Development Company, L.P. Chargeback based storage recommendations for datacenters
US9009204B2 (en) * 2013-04-04 2015-04-14 Hitachi, Ltd. Storage system
US9400792B1 (en) * 2013-06-27 2016-07-26 Emc Corporation File system inline fine grained tiering
JP6164042B2 (ja) * 2013-10-24 2017-07-19 富士通株式会社 中継装置、中継方法、中継プログラム、および中継システム
US9411834B2 (en) 2014-01-14 2016-08-09 Netapp, Inc. Method and system for monitoring and analyzing quality of service in a storage system
US9542346B2 (en) 2014-01-14 2017-01-10 Netapp, Inc. Method and system for monitoring and analyzing quality of service in a storage system
US9658778B2 (en) 2014-01-14 2017-05-23 Netapp, Inc. Method and system for monitoring and analyzing quality of service in a metro-cluster
US9547445B2 (en) 2014-01-14 2017-01-17 Netapp, Inc. Method and system for monitoring and analyzing quality of service in a storage system
US9542103B2 (en) 2014-01-14 2017-01-10 Netapp, Inc. Method and system for monitoring and analyzing quality of service in a storage system
US10853365B2 (en) * 2014-03-28 2020-12-01 Hitachi, Ltd. Database management system, computer, and database management method
US9400724B2 (en) * 2014-05-29 2016-07-26 Dell Products, Lp Federating and protecting multiple big data sources combined with enterprise data sources for BI
CN105376285A (zh) * 2014-08-29 2016-03-02 纬创资通股份有限公司 网络储存去重方法及服务器
CN104239518B (zh) 2014-09-17 2017-09-29 华为技术有限公司 重复数据删除方法和装置
US9489137B2 (en) * 2015-02-05 2016-11-08 Formation Data Systems, Inc. Dynamic storage tiering based on performance SLAs
US9733836B1 (en) * 2015-02-11 2017-08-15 Violin Memory Inc. System and method for granular deduplication
US9824092B2 (en) 2015-06-16 2017-11-21 Microsoft Technology Licensing, Llc File storage system including tiers
US10628391B1 (en) * 2015-09-30 2020-04-21 EMC IP Holding Company LLC Method and system for reducing metadata overhead in a two-tier storage architecture
US10698606B2 (en) * 2016-08-30 2020-06-30 Lenovo Enterprise Solution (Singapore) Pte. Ltd. Host aware storage tiering
US10613988B2 (en) * 2016-09-28 2020-04-07 Micro Focus Llc Purging storage partitions of databases
US10678431B1 (en) * 2016-09-29 2020-06-09 EMC IP Holding Company LLC System and method for intelligent data movements between non-deduplicated and deduplicated tiers in a primary storage array
US10162531B2 (en) * 2017-01-21 2018-12-25 International Business Machines Corporation Physical allocation unit optimization
US11886732B2 (en) 2017-01-31 2024-01-30 Seagate Technology Llc Data storage server with multi-memory migration
JP6878369B2 (ja) * 2018-09-03 2021-05-26 株式会社日立製作所 ボリューム配置管理装置、ボリューム配置管理方法、及びボリューム配置管理プログラム
SG11202002775RA (en) * 2019-09-12 2020-04-29 Alibaba Group Holding Ltd Log-structured storage systems
EP3669263B1 (fr) * 2019-09-12 2022-03-02 Advanced New Technologies Co., Ltd. Systèmes de stockage structurés par journalisation
WO2019228574A2 (fr) 2019-09-12 2019-12-05 Alibaba Group Holding Limited Systèmes de stockage structuré par journaux
WO2019228569A2 (fr) 2019-09-12 2019-12-05 Alibaba Group Holding Limited Systèmes de stockage à structure de journal
WO2019228571A2 (fr) 2019-09-12 2019-12-05 Alibaba Group Holding Limited Systèmes de stockage à structure journalisée
US10942852B1 (en) 2019-09-12 2021-03-09 Advanced New Technologies Co., Ltd. Log-structured storage systems
SG11202002614XA (en) 2019-09-12 2020-04-29 Alibaba Group Holding Ltd Log-structured storage systems
CN111183420B (zh) 2019-09-12 2023-06-30 创新先进技术有限公司 日志结构存储系统
SG11202002587TA (en) 2019-09-12 2020-04-29 Alibaba Group Holding Ltd Log-structured storage systems
US20220245104A1 (en) * 2021-02-02 2022-08-04 Maxlinear, Inc. Hashing for deduplication through skipping selected data
US20230401176A1 (en) * 2022-06-14 2023-12-14 Rubrik, Inc. Storage tiering for computing system snapshots

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH0342A (ja) 1989-05-29 1991-01-07 Japan Medical Dynamic Marketing Inc 頭蓋内圧測定装置
EP1830270A1 (fr) * 2006-03-01 2007-09-05 Quantum Corporation Système de stockage de données comportant un gestionnaire de piscine pour bloc unique et applications de stockage étagé
US20090319585A1 (en) * 2008-06-24 2009-12-24 Parag Gokhale Application-aware and remote single instance data management
US7822939B1 (en) * 2007-09-25 2010-10-26 Emc Corporation Data de-duplication using thin provisioning
US20100332401A1 (en) * 2009-06-30 2010-12-30 Anand Prahlad Performing data storage operations with a cloud storage environment, including automatically selecting among multiple cloud storage sites

Family Cites Families (25)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5325509A (en) 1991-03-05 1994-06-28 Zitel Corporation Method of operating a cache memory including determining desirability of cache ahead or cache behind based on a number of available I/O operations
US5636355A (en) 1993-06-30 1997-06-03 Digital Equipment Corporation Disk cache management techniques using non-volatile storage
JPH086854A (ja) * 1993-12-23 1996-01-12 Unisys Corp アウトボードファイルキャッシュ外部処理コンプレックス
US6574635B2 (en) 1999-03-03 2003-06-03 Siebel Systems, Inc. Application instantiation based upon attributes and values stored in a meta data repository, including tiering of application layers objects and components
US6839803B1 (en) 1999-10-27 2005-01-04 Shutterfly, Inc. Multi-tier data storage system
JP2001166993A (ja) 1999-12-13 2001-06-22 Hitachi Ltd 記憶制御装置およびキャッシュメモリの制御方法
US7237021B2 (en) 2003-04-04 2007-06-26 Bluearc Uk Limited Network-attached storage system, device, and method supporting multiple storage device types
EP1668486A2 (fr) 2003-08-14 2006-06-14 Compellent Technologies Systeme et procede de lecteur de disque virtuel
US7543296B2 (en) 2003-08-26 2009-06-02 International Business Machines Corporation Time based multi-tiered management of resource systems
US7441096B2 (en) 2004-07-07 2008-10-21 Hitachi, Ltd. Hierarchical storage management system
US7062624B2 (en) 2004-09-29 2006-06-13 Hitachi, Ltd. Method for managing volume groups considering storage tiers
US7606934B1 (en) 2005-03-10 2009-10-20 Sun Microsystems, Inc. Dynamic routing of I/O requests in a multi-tier storage environment
US7539709B1 (en) 2005-06-15 2009-05-26 Sun Microsystems, Inc. Dynamic data migration in a multi-tier storage system
US7613876B2 (en) 2006-06-08 2009-11-03 Bitmicro Networks, Inc. Hybrid multi-tiered caching storage system
US7805420B2 (en) 2006-11-20 2010-09-28 Microsoft Corporation Versioning and concurrency control for multiple client access of data
US7734954B2 (en) 2007-01-03 2010-06-08 International Business Machines Corporation Method, computer program product, and system for providing a multi-tiered snapshot of virtual disks
US8001327B2 (en) 2007-01-19 2011-08-16 Hitachi, Ltd. Method and apparatus for managing placement of data in a tiered storage system
JP4922834B2 (ja) * 2007-05-29 2012-04-25 株式会社日立製作所 コンピュータシステムに存在するリソースの性能を監視する装置及び方法
US8224782B2 (en) 2008-09-29 2012-07-17 Hitachi, Ltd. System and method for chunk based tiered storage volume migration
US8566549B1 (en) * 2008-12-31 2013-10-22 Emc Corporation Synchronizing performance requirements across multiple storage platforms
US20100199036A1 (en) 2009-02-02 2010-08-05 Atrato, Inc. Systems and methods for block-level management of tiered storage
US8880835B2 (en) 2009-07-07 2014-11-04 International Business Machines Corporation Adjusting location of tiered storage residence based on usage patterns
US8452941B2 (en) 2009-09-23 2013-05-28 Dell Products L.P. System and method for distributing random and sequential data in a tiered storage system
US8375180B2 (en) * 2010-02-05 2013-02-12 International Business Machines Corporation Storage application performance matching
US9213731B2 (en) 2010-05-13 2015-12-15 Symantec Corporation Determining whether to relocate data to a different tier in a multi-tier storage system

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH0342A (ja) 1989-05-29 1991-01-07 Japan Medical Dynamic Marketing Inc 頭蓋内圧測定装置
EP1830270A1 (fr) * 2006-03-01 2007-09-05 Quantum Corporation Système de stockage de données comportant un gestionnaire de piscine pour bloc unique et applications de stockage étagé
US7822939B1 (en) * 2007-09-25 2010-10-26 Emc Corporation Data de-duplication using thin provisioning
US20090319585A1 (en) * 2008-06-24 2009-12-24 Parag Gokhale Application-aware and remote single instance data management
US20100332401A1 (en) * 2009-06-30 2010-12-30 Anand Prahlad Performing data storage operations with a cloud storage environment, including automatically selecting among multiple cloud storage sites

Also Published As

Publication number Publication date
US20190278719A1 (en) 2019-09-12
US10853274B2 (en) 2020-12-01
US10216651B2 (en) 2019-02-26
US20130117515A1 (en) 2013-05-09

Similar Documents

Publication Publication Date Title
US10853274B2 (en) Primary data storage system with data tiering
US10891054B2 (en) Primary data storage system with quality of service
US20200019516A1 (en) Primary Data Storage System with Staged Deduplication
US10275351B2 (en) Journal management
EP3230864B1 (fr) Alerte de nuage à des reproductions
US9720601B2 (en) Load balancing technique for a storage array
US10359967B2 (en) Computer system
US11880578B2 (en) Composite aggregate architecture
US11327653B2 (en) Drive box, storage system and data transfer method
US11249666B2 (en) Storage control apparatus
US11775175B2 (en) Storage system, storage management method, and storage management program

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 12791916

Country of ref document: EP

Kind code of ref document: A1

DPE1 Request for preliminary examination filed after expiration of 19th month from priority date (pct application filed from 20040101)
NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 12791916

Country of ref document: EP

Kind code of ref document: A1